DRAFT

A.R.A.P Report No. 725

PC-SCIPUFF Version 1.3

Technical Documentation
R. Ian Sykes, et al.

Titan Corporation
ARAP Group
P.O. Box 2229
Princeton, NJ 08543-2229

December 2000
Technical Report

CONTRACT No. DNA 001-95-C-0180

DRAFT

DRAFT

ii

DRAFT
SUMMARY
Version 1.3 of the Second-order Closure Integrated Puff model, SCIPUFF, is
described. SCIPUFF is an atmospheric dispersion model with a wide range of
application. The turbulent diffusion parameterization is based on second-order
turbulence closure theory, which relates the dispersion rate to velocity fluctuation
statistics. In addition to the average concentration value, the closure model provides a
prediction of the statistical variance in the concentration field resulting from the random
fluctuations in the wind field. The variance is used to estimate a probability distribution
for the predicted value.
SCIPUFF uses a collection of Gaussian puffs to represent an arbitrary threedimensional, time-dependent concentration field, and incorporates an efficient scheme for
splitting and merging puffs. Wind shear effects are accurately modeled, and puffs are
split when they grow too large for single point meteorology to be representative. These
techniques allow the puff model to describe complex flow effects on dispersion, such as
terrain-driven circulations.
SCIPUFF has been developed with a flexible interface, to describe many types of
source geometries and material properties. Solid particles, liquid droplets, and gaseous
materials are represented, with both primary and secondary evaporation mechanisms that
produce vapor puffs as the droplets evaporate in the air or after deposition on the ground.
Precipitation washout effects are also included for particles and droplets. SCIPUFF
describes dynamic effects of buoyant rise due to thermal release or lighter-than-air
materials, and also the effects of a dense cloud near the ground surface.
The model also uses several types of meteorological input, including surface and
upper air observations or three-dimensional grid data. Planetary boundary layer
turbulence is represented explicitly in terms of surface heat flux and shear stress using
parameterized profile shapes.
A number of enhancements and improvements to Version 1.0 are described.
These include the capability to estimate the uncertainties in the meteorological inputs,
due to forecast or extrapolation errors (see Sections 9.4 and 12.4). A model for flashing
liquid/aerosol effects associated with the instantaneous release of volatile liquids stored
iii

DRAFT
under pressure is included (see Section 6.3). Casualty tables are available for nuclear
weapon releases, using a world-wide population database (see Section 9.5). Other
enhancements include a generalized description of the radiation dose from nuclear
facility releases (Section 8.2), and a vegetative/urban canopy model for transport and
dispersion (Section 11.4).
The technical basis of SCIPUFF and the revised structure of the input and output
files are described in this document. A section describing a number of test calculations
and comparison with observational data is also included (see Section 14).

iv

DRAFT

DRAFT
TABLE OF CONTENTS
Section

Page

SUMMARY .................................................................................................... III

1. INTRODUCTION ....................................................................................... 1
2. TRANSPORT AND DIFFUSION ............................................................... 3
2.1 PUFF MOMENT EQUATIONS. ..................................................... 3
2.2 CONCENTRATION FLUCTUATION VARIANCE. ...................... 12
2.3 SURFACE DEPOSITION................................................................. 25
2.4 SURFACE INTEGRALS. ................................................................ 26
3. NUMERICAL TECHNIQUES .................................................................... 28
3.1 PUFF SPLITTING SCHEME. ......................................................... 29
3.2 PUFF MERGING SCHEME. ........................................................... 33
3.3 PUFF INVERSION CAPPING. ....................................................... 37
3.4 FINITE-DIFFERENCE SCHEMES. ................................................ 38
3.5 ADAPTIVE TIME STEPS............................................................... 41
3.6 STATIC PUFFS............................................................................... 43
3.7 ADAPTIVE SURFACE GRIDS....................................................... 46
4. PARTICLE MATERIALS........................................................................... 48
4.1 PARTICLE MATERIAL PROPERTIES.......................................... 48
4.2 PARTICLE GRAVITATIONAL SETTLING. .................................. 48
4.3 DRY DEPOSITION......................................................................... 51
4.4 PRECIPITATION WASHOUT........................................................ 56
4.5 EFFECTS ON DIFFUSION. ............................................................ 61
5. LIQUID MATERIALS ................................................................................ 63
5.1 LIQUID SPECIFICATIONS............................................................ 63
5.2 PRIMARY EVAPORATION........................................................... 64
5.3 SECONDARY EVAPORATION...................................................... 71

vi

DRAFT
TABLE OF CONTENTS (Continued)
Section

Page

6. DYNAMIC EFFECTS................................................................................. 77
6.1 BUOYANCY AND MOMENTUM RISE. ....................................... 77
6.2 DENSE GAS EFFECTS................................................................... 83
6.3 FLASHING LIQUID/AEROSOL EFFECTS.................................... 91
7. SOURCE SPECIFICATION........................................................................ 98
7.1 CONTINUOUS SOURCES. ............................................................ 98
7.2 INSTANTANEOUS SOURCES. ..................................................... 99
7.3 LIQUID POOL EVAPORATION. ................................................... 100
7.4 FLASHING LIQUID/AEROSOL SOURCES. ................................. 102
8. SPECIAL MATERIALS.............................................................................. 104
8.1 ACTIVITY DECAY. ....................................................................... 104
8.2 NUCLEAR WEAPON MATERIALS. ............................................. 105
8.3 NUCLEAR FACILITY MATERIALS. ............................................ 107
9. MODEL OUTPUT ...................................................................................... 112
9.1 LOCAL CONCENTRATION VALUES. ......................................... 112
9.2 PROBABILISTIC CALCULATION................................................ 114
9.3 POPULATION/AREA EXPOSURE. ............................................... 115
9.4 HAZARD AREAS. .......................................................................... 117
9.5 CASUALTY OUTPUT. ................................................................... 118
10. METEOROLOGY SPECIFICATION........................................................ 124
10.1 BACKGROUND............................................................................ 124
10.2 OBSERVATIONAL INPUT. ......................................................... 124
10.3 GRIDDED INPUT. ........................................................................ 129
10.4 MEAN FIELD INTERPOLATION. ............................................... 130
10.5 MASS-CONSISTENCY. ............................................................... 131

vii

DRAFT
TABLE OF CONTENTS (Continued)
Section

Page

11. PLANETARY BOUNDARY LAYER....................................................... 134

11.1 GOVERNING PARAMETERS. .................................................... 134
11.2 SPECIFYING PBL PARAMETERS. ............................................. 137
11.3 PBL MEAN PROFILES................................................................. 147
11.4 VEGETATIVE/URBAN CANOPY PROFILES............................. 149
12. TURBULENCE SPECIFICATION ........................................................... 152
12.1 PLANETARY BOUNDARY LAYER. .......................................... 152
12.2 VEGETATIVE/URBAN CANOPY LAYER.................................. 154
12.3 MESOSCALE/SYNOPTIC SCALE............................................... 155
12.4 METEOROLOGICAL UNCERTAINTIES.................................... 158
13. FILE FORMATS ....................................................................................... 164
13.1 INPUT FILES. ............................................................................... 164
13.2 METEOROLOGY INPUT. ............................................................ 185
13.3 OUTPUT FILES. ........................................................................... 201

14. MODEL TEST CALCULATIONS ............................................................ ERROR! BOOKMAR

14.1 SHORT RANGE DIFFUSION....................................................... ERROR! BOOKMAR

14.2 LONG RANGE DIFFUSION......................................................... ERROR! BOOKMAR

14.3 DYNAMIC RISE EFFECTS. ......................................................... ERROR! BOOKMAR

14.4 DENSE GAS EFFECTS................................................................. ERROR! BOOKMAR

15. REFERENCES .......................................................................................... 236

viii

DRAFT
FIGURES
Figure
2-1

Page
Normalized fluctuation dissipation timescale as a function of
dimensionless time for a shear-distorted Gaussian puff. The solid line is
the analytic solution (2.50). Dashed line is the model prediction from
(2.51) and (2.52)............................................................................................. 18

3-1

Maximum dimensionless error, s, from splitting a single Gaussian puff

as a function of separation distance, s. The error is relative to the
maximum value in the original Gaussian, and the separation distance of
the puffs after splitting is relative to the original Gaussian spread. Results
are shown for both one-dimensional and two-dimensional splits..................... 32

3-2

Schematic illustration of puff reflection for a split below the ground

surface............................................................................................................ 32

3-3

Schematic illustration of the adaptive multi-grid for locating puffs. Grid

cell numbers are represented as C-n, and two levels of refinement are
shown. See text for a description of the cell storage rules .............................. 35
3-4

Schematic illustration of the use of static puffs to represent a continuous

source. The shaded puff is stored as the release description at the end of
the static phase ............................................................................................... 44

4-1

Scavenging coefficient versus particle diameter. Solid lines denote the

Seinfeld (1986) model described above and the dashed lines denote the
model of Nieto et al. (1994) (with a Marshall and Palmer raindrop size
distribution). The precipitation rate ( Pr ) is equal to 0.5 and 25 mmh1 for
drizzle and heavy rain, respectively. ............................................................... 59

4-2

Scavenging coefficient versus particle diameter. Solid lines denote the

model implemented in SCIPUFF and the dashed lines denote the model
of Horn et al. (1988). The precipitation rate ( Pr ) is equal to 0.5 and 10

mmh1 for drizzle and moderate rain, respectively .......................................... 60

ix

DRAFT
FIGURES (Continued)
Figure

Page

5-1

Terminal velocity, Sherwood (Sh), and Nusselt (Nu) numbers for agent
VX. Solid line is SCIPUFF algorithm, short dash lines are from
NUSSE4. Left column (a) is at 100m altitude, column (b) is at 10km ............ 67

5-2

Evolution of droplet diameter (D), altitude (H), fall velocity (V) and
droplet temperature (T) for GB droplets released at a height of 10km with
different initial diameters. Solid line is SCIPUFF, dashed line is
NUSSE4......................................................................................................... 68

6-1

Schematic illustration of slumping dense cloud. ............................................. 85

6-2

Simplified velocity field representation for a dense cloud. .............................. 85

11-1 Schematic illustration of the idealized velocity profile in the planetary

boundary layer................................................................................................ 149
11-2 Canopy velocity profile shapes from (11.47) and (11.48) compared with
the experimental data of Cionco (1972). ......................................................... 151
12-1 Turbulence profiles within the canopy layer for different values of the
canopy parameter, c . (a) horizontal turbulence (shear driven
component), (b) vertical turbulence component. ............................................. 156
12-2 Normalized error variance growth for forecast and persistence
assumptions.................................................................................................... 160
131 Sample PC-SCIPUFF input file ...................................................................... 165
13-2 Sample PC-SCIPUFF release scenario file...................................................... 173
13-3 Sample PC-SCIPUFF meteorology scenario file............................................. 176
13-4 Example PC-SCIPUFF sampler location file .................................................. 179

DRAFT
FIGURES (Continued)
Figure

Page

13-5 General structure of a RAD file. Keywords are in upper case; lower case
italics indicate user-specified input. Brackets denote optional input ............... 182
13-6 Header structure for a meteorology observation file. Records and
variables in brackets are optional or required only if nvarp is given ............... 185
13-7 General structure of numerical data in a meteorological observation file.
(a) Without nvarp specified; (b) with nvarp specified ................................... 193
13-8 Fortran pseudo-code for reading a formatted MEDOC input file..................... 195
13-9 General structure of an HPAC gridded meteorology file ................................. 197
13-10 General structure of a terrain file .................................................................... 199
13-11 Fortran statements to write a surface output file header (HPAC3.2 only) ........ 202
13-12 Fortran statements to write a surface output file timebreak ............................. 203
13-13 Fortran statements to calculate field values from the surface output file.......... 205
13-14 Example of a sampler output file .................................................................... 209
14-1 Comparison between SCIPUFF predictions (solid lines) and the PGT
dispersion curves for categories A through F .................................................. 211
14-2 Comparison between the observed maximum concentrations and
SCIPUFF predictions for the passive releases in the Model Data Archive.
Various experiments are indicated by the letter codes ..................................... 214
14-3 Comparison between SCIPUFF predictions and the observational data of
Weil et al. (1993)............................................................................................ 215
14-4 Comparison between SCIPUFF predictions and the observational data of
Mikkelsen et al. (1988)................................................................................... 216

xi

DRAFT
FIGURES (Continued)
Figure

Page

14-5 Comparison between SCIPUFF predictions and the observational data of

Hgstrm (1964) ............................................................................................ 217
14-6 Comparison between SCIPUFF predictions of mean concentration and
the observational data from CONFLUX 1994................................................. 219
14-7 Comparison between SCIPUFF predictions of standard deviation to mean
concentration ratio and the observational data of CONFLUX 1994................. 220
14-8 Comparison of mean concentration between SCIPUFF predictions and
ETEX data (Mosca et al., 1997). Contours are 0.01, 0.02, 0.05, 0.1, 0.2,
0.5, 1, 2, 5, 10, 20, and 50 ngm3 .................................................................... 222
14-9 Comparison between 3-hr average concentrations from SCIPUFF
predictions (dashed) and ETEX data (solid line) at 9 selected locations
(Mosca et al., 1997). Concentration is in ngm3 ............................................. 223
14-10 Comparison between total integrated dose from SCIPUFF predictions and
ETEX data (solid line) at all locations (Mosca et al., 1997). Dose is in ng
s m3. (a) compares each station, (b) shows ordered data, unmatched by
location. ......................................................................................................... 226
14-11 Momentum (non-buoyant) jet centerline height, z , as a function of
downstream distance for a range of R. Symbols are the data of Gordier
(1959) - from Hirst (1971), solid lines are the SCIPUFF predictions............... 228
14-12 Comparison between SCIPUFF predictions of momentum jet centerline
heights for a range of R (solid lines) and the one-third law, (13.3), with
two values of (short dashes, = 0.6; long dashes, = 0.5)........................... 228
14-13 Comparison between SCIPUFF predictions for buoyancy-dominated jet
centerline heights (solid lines) and theory given by (13.5) (dashed lines).
4
4
Case A: R = 10, Fb2 U a Fm = 0.01; Case B: R = 10, Fb2 U a Fm = 1 ; Case
C: R = 1, Fb2 U a4 Fm = 1 ................................................................................. 229
xii

DRAFT
FIGURES (Continued)
Figure

Page

14-14 Comparison of SCIPUFF buoyancy-dominated jet centerline heights in a

uniformly stratified background (solid lines) with (13.6) (dashed lines).
4
For all cases, R = 10. Case A: Fb2 U a Fm = 0.01, Fb N U a4 = 0.00572 ;
Case B: Fb2 U a4 Fm = 1 , Fb N U a4 = 0.00572 ; Case C: Fb2 U a4 Fm = 1 ,
Fb N U a4 = 0.181............................................................................................. 230
14-15 Centroid height of a light bubble released into a neutral quiescent
background. Symbols are the data of Mantrom and Haigh (1973), solid
lines are the SCIPUFF predictions. The Reynolds number for the
experiments are based on the initial bubble diameter and the terminal
velocity of a corresponding non-entraining sphere. Case A is for a single
puff release with a Gaussian spread of D0/2; Case B is for a release
comprised of 136 puffs representing a uniform distribution of mass within
a spherical bubble of diameter D0. (a) Early time behavior; (b) expanded
scale showing late time behavior .................................................................... 232
14-16 Comparison between the observed maximum concentrations and
SCIPUFF predictions for the dense gas releases in the Model Data
Archive. Various experiments are indicated by the letter codes...................... 234

xiii

DRAFT
TABLES
Table

4-1

Page

Precipitation rates in mm/hr associated with precipitation types...................... 61

11-1 Suggested values for surface roughness (Saucier, 1987). ................................ 139
11-2 Relationship between stability index, PGT stability class, MoninObukhov length, L, and an assumed boundary layer depth, z i , if not
specified as input............................................................................................ 140
11-3 Suggested values for surface albedo as a function of land-use and season
(Paine, 1987). ................................................................................................. 143
11-4 Suggested values for Bowen ratio as a function of land-use and season
(Paine, 1987). ................................................................................................. 143
131 Description of recognized keywords and associated input for RAD files. ........ 183
132 Valid keywords and associated input strings for PROFILE files. .................... 186
133 Meteorological variables and their units recognized in SCIPUFF. .................. 189
134 Valid landuse indices and associated landuse types for terrain files................. 200
141 Model performance statistics for SCIPUFF prediction of the ETEX
sampler concentrations for different time periods. .......................................... 225

xiv

DRAFT
SECTION 1
INTRODUCTION
SCIPUFF is a Lagrangian transport and diffusion model for atmospheric
dispersion applications. The acronym SCIPUFF stands for Second-order Closure
Integrated PUFF and describes two basic aspects of the model. First, the numerical
technique employed to solve the dispersion model equations is the Gaussian puff method
(Bass, 1980) in which a collection of three-dimensional puffs is used to represent an
arbitrary time-dependent concentration field.
Second, the turbulent diffusion
parameterization used in SCIPUFF is based on the second-order turbulence closure
theories of Donaldson (1973) and Lewellen (1977), providing a direct connection
between measurable velocity statistics and the predicted dispersion rates.
The Lagrangian puff methodology affords a number of advantages for
atmospheric dispersion applications from localized sources. The Lagrangian scheme
avoids the artificial diffusion problems inherent in any Eulerian advection scheme, and
allows an accurate treatment of the wide range of length scales as a plume or cloud grows
from a small source size and spreads onto larger atmospheric scales. This range may
extend from a few meters up to continental or global scales of thousands of kilometers.
In addition, the puff method provides a very robust prediction under coarse resolution
conditions, giving a flexible model for rapid assessment when detailed results are not
required. The model is highly efficient for multiscale dispersion problems, since puffs
can be merged as they grow and resolution is therefore adapted to each stage of the
diffusion process.
The efficiency of SCIPUFF has been improved by the implementation of adaptive
time stepping and output grids. Each puff uses a time step appropriate for resolving its
local evolution rate, so that the multiscale range can be accurately described in the time
domain without using a small step for the entire calculation. The output spatial fields are
also computed on an adaptive grid, avoiding the need for the user to specify grid
information and providing a complete description of the concentration field within the
computational constraints under most conditions.
The generality of the turbulence closure relations provides a dispersion
representation for arbitrary conditions. Empirical models based on specific dispersion
1

DRAFT
data are limited in their range of application, but the fundamental relationship between
the turbulent diffusion and the velocity fluctuation statistics is applicable for a much
wider range. Our understanding of the daytime planetary boundary layer velocity
fluctuations provides reliable input for the second-order closure description of dispersion
for these conditions. For larger scales and upper atmosphere stable conditions, the
turbulence description is based on climatological information, but the closure framework
is in place to accept improvement as our understanding of these regimes improves. The
closure model has been applied on local scales up to 50km range (Sykes et al., 1988) and
also on continental scales up to 3000km range (Sykes et al., 1993c).
The second-order closure model also provides the probabilistic feature of
SCIPUFF through the prediction of the concentration fluctuation variance. In addition to
giving a mean value for the concentration field, SCIPUFF provides a quantitative value
for the random variation in the concentration value due to the stochastic nature of the
turbulent diffusion process. This uncertainty estimate is used to provide a probabilistic
description of the dispersion result, and gives a quantitative characterization of the
reliability of the prediction. For many dispersion calculations, the prediction is inherently
uncertain due to a lack of detailed knowledge of the wind field and a probabilistic
description is the only meaningful approach.
The following sections will describe the technical basis of the SCIPUFF
prediction, as well as the treatment of additional physical phenomena, such as particle
deposition rates. The current model can treat multiple materials, including gases,
particles, and liquid droplets, and multiple sources, both continuous plumes and
instantaneous puffs.

DRAFT
SECTION 2
TRANSPORT AND DIFFUSION
2.1 PUFF MOMENT EQUATIONS.
2.1.1 Gaussian Moment Definition.

SCIPUFF uses a Gaussian puff representation for the concentration field of a

dispersing contaminant. In this section, we will restrict discussion to gaseous materials;
solid particles and liquid materials are discussed in Sections 4 and 5. A threedimensional Gaussian is completely described by its spatial integral moments up to
second-order, and can be written in the form


Q
1
exp ij1 x i x i x j x j
1/ 2
2
Det ( )

3/ 2

cx =

(2.1)

For general atmospheric dispersion problems, we must also consider the effects of the
ground surface and the capping inversion at the top of the planetary boundary layer, which
are usually represented as reflective surfaces. The reflection description for the
generalized Gaussian will be described in Section 9.1. For our present discussion, we
restrict attention to the moment equations, which are independent of the shape assumption.
Using an angle bracket to denote an integral over all space, the spatial moments in
(2.1) are defined as

Zeroth moment - mass

Q= c

(2.2)

First moment - centroid

Q xi = c xi

(2.3)

Second moment - spread

Q ij = c x i x i x j x j

gc

(2.4)

DRAFT
Note that this generalized tensor definition of the moments is completely
independent of any specific coordinate system, and can therefore describe an arbitrarily
oriented Gaussian shape. The full moment description requires 10 numbers to represent
the mass, the 3 centroid coordinates, and 6 spread moments since ij is a symmetric
second-rank tensor. The moments describe the unreflected Gaussian shape, but local
concentration values must account for surface reflection as described in Section 9.1. The
evaluation of the ambient meteorological variables should also strictly use the true
centroid location of the reflected puff, but this involves complex calculations. We
therefore simply use the larger of z and 33 as the height (above the ground surface) at
which the meteorological field variables, such as velocity and turbulence, are evaluated.

The specific Gaussian variation (2.1) applies to an individual puff, but in general
the local concentration field will be composed of a sum of contributions from a number
of such puffs. The details of the concentration calculation are described in Section 9.1.
In this section, we describe the transport and diffusion model equations for the individual
puff moments.

2.1.2 Moment Transport Equations.

The advection-diffusion equation for a scalar quantity in an incompressible flow

field can be written as

+
( ui c ) = k 2 c + S
t xi

(2.5)

a f

where ui x, t is the turbulent velocity field, k is the molecular diffusivity, and S represents

the source terms. Surface deposition, described in Section 2.3, is an example of a source
term, but other terms associated with different materials are described in Sections 4 and 5.
The atmospheric velocity field is generally turbulent, so we use the Reynolds
averaging technique to define a mean and a turbulent fluctuation value. Denoting the
mean by an overbar and the fluctuation by a prime, we have u = u + u , and a similar
decomposition can be applied to the concentration, c. The Reynolds averaged
conservation equation for the mean scalar concentration is thus

+
uic + k 2 c + S
( ui c ) =
t xi
xi

( )

(2.6)

DRAFT
where uic is the turbulent concentration flux.
At this point, the averaging in the overbar involves an unspecified ensemble of
velocity fields. In some cases, the ensemble may be considered to include the
conventional planetary boundary layer turbulence only, and the mean wind, u , is defined
in the usual sense. The dispersion framework developed here is more general, however,
and can include larger scale variability, or may be more restrictive and ignore meandering
motions. The specification of the turbulent fluctuation values will be discussed more
fully in Section 12.
Neglecting source terms, S, for the present, (2.6) is readily integrated over the
spatial dimensions to give conservation of mass
dQ
=0
dt

(2.7)

where Q now represents the integrated mean concentration, c . Note that (2.7) applies to
an inert conserved tracer, with no loss or transformation processes; these effects will be
described later.
Equations for the higher spatial moments of c are obtained similarly, but require
some assumption about the spatial variation of the velocity field. The simplest
assumption is a constant velocity based on the value at the puff centroid, but this neglects
any effect of shear. We therefore use a linear representation for the local velocity field
ui ( x ) = ui ( x ) + ( x j x j )

ui
xj

(2.8)

where the velocity gradient is also evaluated at the centroid location.

Multiplying (2.6) by xi and integrating by parts gives
uic
uc
d xi
= i +
dt
Q
Q

(2.9)

where the first term on the right-hand side represents transport by the mean wind and the
second term is the turbulent drift. The latter arises from correlation between the velocity
fluctuations and the concentration fluctuations. Substituting for u from (2.8) and
assuming any symmetric spatial distribution for c, the velocity gradient term vanishes to
give

DRAFT
uic
dx i
= ui x +
dt
Q

af

(2.10)

The second-moment equation is obtained by multiplying (2.6) by x ix , where

j
x i = x i x i , and integrating by parts to give
d ij
dt

= ik

u j
xk

+ jk

xiu j c
x j uic
ui
+
+
xk
Q
Q

(2.11)

where the first two terms represent the shear distortion of the puff and the second two
terms represents the turbulent diffusion. The pairs of terms are required for the
ij-symmetry of the second moment tensor.
This completes the specification of the moment equations in the absence of
external source terms, S, but we have introduced turbulent flux moments in (2.10) and
(2.11). This is the essence of the turbulence closure problem, in which the Reynolds
average operator always introduces higher-order fluctuation correlations. The turbulent
fluctuations must be modeled empirically at some level and we next discuss the
turbulence closure technique used in SCIPUFF.

2.1.3 Turbulence Closure Diffusion Model.

2.1.3.1 Second-order Closure Framework. The diffusion model in SCIPUFF is based on

second-order turbulence closure, which provides a transport equation for the secondorder fluctuation terms. First-order closure prescribes the turbulent fluxes in terms of the
local mean gradients using an empirical turbulent diffusivity, but a more general relation
can be obtained from a higher-order closure. It is not our intention to review turbulence
closure theory here, we only provide the basic model description; the interested reader
can find detailed discussions in the literature, e.g., Mellor and Herring (1973), Launder et
al. (1975), Lewellen (1977).
The equations for the puff moments involve the turbulent flux of concentration,
and a rigorous conservation equation can be derived for this quantity from the scalar and
momentum equations. Neglecting the molecular diffusion terms, the flux transport
equation can be written in the form

DRAFT
u

p gi
uic + u j
uic = uiu j
u j c i
uiu j c c
+ c
t
xj
xj
xj xj
xi T0

(2.12)

This equation involves higher-order terms, such as the triple correlation and the pressure
correlation, which must be modeled empirically. We use the model of Lewellen (1977) to
form a closed equation for the turbulent flux, giving

u c i +
v c q
uic
uic + u j
uic = uiu
j
j
t
x j
x j
x j x j
x j

GH

JK

I
(2.13)

g
q
uic + i c
T0

where A = 0.75 and vc = 0.3 are empirical model constants. The turbulent velocity scale,
q, is defined as q 2 = uiui , and
is the turbulent length scale; these quantities will be
discussed more fully below. The buoyancy term uses the Boussinesq approximation, and
gi = 0,0, g is the gravitational acceleration, T0 is the reference temperature, and is the
potential temperature fluctuation.

The general equation (2.13) for the turbulent fluxes can be integrated spatially to
provide transport equations for the flux moments in (2.10) and (2.11). However, the
multiple tensor indices give a large number of correlations to consider in the general case.
We therefore introduce some restrictions for the atmospheric dispersion cases that allow
us to neglect many of the correlations. First, we only consider the vertical component of
the turbulent drift, uic , since advection by the mean wind will generally dominate in
the horizontal directions. Second, the only off-diagonal component of the flux moment
tensor, x i u c , to be considered will be the symmetric horizontal term
j
X12 = x1 u 2 c + x 2 u1c

(2.14)

In general, the off-diagonal terms represent puff distortions due to velocity covariances.
Such covariances are usually generated by wind shear, which will dominate the distortion
process through the mean shear terms in (2.11). The horizontal distortion is retained for
the special case of large-scale horizontal fluctuations, as described in Section 12.3. With
these restrictions, we now describe the horizontal and vertical diffusion representations.

2.1.3.2 Horizontal Diffusion. Equations for the two diagonal horizontal moments are
derived directly from (2.13) as

DRAFT
d
q
x u c = Qu 2 A
x u c
dt

(2.15)

d
q
y v c = Q v 2 A
y v c
dt

(2.16)

and

where we have used the component values for the velocity and position vectors, i.e.,
x i = x , y , z and ui = u , v , w .

Under steady homogeneous conditions, (2.16) gives a very simple prediction for
the effective horizontal diffusivity and lateral spread. The diffusivity in the y-direction
can be defined as
Ky =

y v c

(2.17)

Then, if K y = 0 at t = 0, the solution for K y is

K y = v 2 1 exp t

(2.18)

where the horizontal timescale = Aq . The solution for the lateral spread, 2 = 22 ,
y

a f

is then

2 = 2 + 2v 2 t + exp t
0
y

(2.19)

where 0 is the initial spread.

The solution (2.19) is consistent with Taylor's (1921) exact analytic result for
turbulent diffusion in homogeneous stationary conditions, with an exponential form for
the Lagrangian velocity autocorrelation function and an integral timescale of .

The above discussion uses a single length scale, , to describe the horizontal
turbulence spectrum, but this is insufficient for a proper description of the atmospheric
spectrum. The wide range of length scales demands a more complicated treatment, since
each length scale is associated with a distinct timescale. We consider three distinct
horizontal turbulence components, representing shear-driven turbulence, buoyancydriven turbulence, and a large-scale component representing mesoscale or synoptic scale
velocity fluctuations. The three components will be denoted by subscripts S, B, and L,
respectively. The specification of the three components is discussed in Section 12. The

DRAFT
three components are assumed to be statistically independent, so that the correlations can
simply be summed, giving

y v c = y v c + y v c
S

+ y v c

(2.20)

and a similar relation for the x-component.

We further assume that the boundary layer components are isotropic (in the
horizontal plane), so that we only require a single correlation variable for the S and B
terms. Buoyancy generation is clearly isotropic in the horizontal, but the shear
component implies a preferred direction. It is also well known that the streamwise
turbulence intensity is larger than the lateral component. However, in the presence of a
mean wind shear, we expect the explicit velocity gradient terms in (2.11) to dominate the
spread in the streamwise direction and we therefore simplify the turbulence
representation using the lateral component only.
Evolution equations for the horizontal velocity correlation integrals are exactly
like (2.16), except that the turbulent velocity and length scales represent the specific
component, i.e.,
q
d

y v c = Q v S 2 A S y v c
S
S
dt

(2.21)

Similar equations are used for the B and L components. The total correlation integrals are
then written as

x u c = y v c + y v c
S

y v c = y v c + y v c
S

+ x u c
+ y v c

(2.22)
(2.23)

The large scale component is not assumed to be isotropic, and therefore requires
both x- and y-components, and also the off-diagonal X12 defined by (2.14), which satisfies
q
d
X12 = 2Q u v L A L X12
L
dt

(2.24)

Details of the specification of the velocity fluctuations and length scales are
provided in Section 12.

DRAFT
2.1.3.3 Vertical Diffusion. The vertical diffusion parameterization is more complicated
than the horizontal for two reasons. First, the inhomogeneity is more pronounced in the
vertical direction, so that vertical gradients cannot be ignored, and second, the buoyancy
forces introduce an additional phenomenon in the vertical direction. Dealing first with
the first moment of the vertical flux, z w c , integration of the full second-order closure
expression (2.13) gives
q
d
g
z w c = Q w 2 A V z w c +
z c
dt
V
T0

(2.25)

where the subscript V denotes the vertical component for the turbulence velocity and
scale.
A similar equation can be derived for the temperature correlation, using the model
equations of Lewellen (1977), as

q
d
d
z c = Q w 2 bs V z c
z w c
dt
dz
V

(2.26)

where b = 0.125 and s = 1.8 are turbulence model constants.

Equations (2.25) and (2.26) represent the full second-order closure model for the
vertical flux, but give oscillatory solutions for stable temperature gradients and can lead
to negative values of the diffusivity. A more robust scheme is obtained by relaxing the
vertical flux moment toward its equilibrium value on the appropriate timescale, similar to
the horizontal equation (2.16). The equilibrium (steady-state) solution for the vertical
diffusivity, defined as
Kz =

z wc

(2.27)

is obtained from (2.25) as

K zeq =

V
g
w 2 + Gzeq
AqV
T0

FG
H

IJ
K

(2.28)

where
Gz =

z c

(2.29)

Using a similar assumption for the temperature correlation, we obtain

10

DRAFT

V
eq
w
Kz
2bsqV
z

(2.30)

Gzeq =

so that the equilibrium diffusivity is given by

(2.31)

2
1 g V
1+
2
2 Abs T0 qV z

V
AqV

K zeq =

g V w
T0 qV 2bs

w2 +

A simple equation for the vertical flux moment that satisfies the equilibrium
equation and evolves on the proper timescale is therefore


q
d
z w c = A V QK zeq z w c
V
dt

(2.32)

and this is the model equation employed in SCIPUFF. The use of the late-time
equilibrium value (2.31) in (2.32) over-estimates the early time growth of the diffusivity
under convective conditions, however, since the buoyancy term in (2.25) develops on the
same turbulence time scale as the diffusivity itself. We therefore approximate the
equilibrium diffusivity in (2.32) using (2.28) but with a scaled value for Gzeq . The
buoyancy term is scaled by the ratio of Kz to K zeq , so that the growth of the buoyancy
correlation is represented as being proportional to the growth of the diffusivity.
The zeroth moment of the vertical flux, the turbulent vertical drift, is important in
regions with vertical gradients of the turbulent correlations. The role of the drift term in
the model is to account for non-uniformity in the vertical diffusivity, and the equilibrium
drift velocity must be equal to the gradient of the equilibrium diffusivity. We therefore
define a simplified equation for the flux moment as


q
K zeq
d
wc = A V Q
wc
dt
V
z


(2.33)

It is emphasized that (2.33) is required for consistency with the simplified

diffusivity moment equation (2.32), and is not directly derived from the second-order
closure model. The main purpose of (2.33) is to maintain a uniform concentration
distribution across a well-mixed layer with when puffs are splitting, as described in
Section 3.1. Vertical variations in the turbulence fields lead to differential diffusion rates,
which can distort an initially uniform concentration profile unless the drift term is

11

DRAFT
included. A similar effect is found in stochastic particle dispersion models as discussed
by Thomson (1987).

2.2 CONCENTRATION FLUCTUATION VARIANCE.

2.2.1 Concentration Variance Equation.

The probabilistic aspect of the SCIPUFF dispersion prediction is based on the

second-order closure model for the concentration fluctuation variance. The dispersion of
any species in a turbulent velocity field is a random process since the turbulent
fluctuations are effectively chaotic and cannot be measured or predicted in detail. The
scalar concentration is therefore a stochastic quantity, with a probability distribution that
depends on the distribution of velocity fluctuations. Traditional deterministic estimates
of atmospheric dispersion only provide a single concentration value as a function of
space and time, and this corresponds to the mean value, c , for some definition of the
statistical ensemble. The mean value is the first moment of the probability distribution,
and contains no information about the statistical variability in the prediction. Higher
moments are required to give a quantitative description of the variability. The
probabilistic aspect of the SCIPUFF dispersion prediction is based on a transport
equation for the statistical variance in the concentration value, that is the second moment
of the probability distribution. The transport equation is based on the same second-order
turbulence closure theory described in Section 2.1 for the mean dispersion rates. The
mean and the variance are then used to provide a probabilistic prediction using a
parameterized probability density function, described in Section 9.3.
The concentration fluctuation variance equation can be obtained from the scalar
mass conservation equation (2.5) in the form

c
c 2

+
ui c 2 = 2 uic

uic 2 2k
xi
t
xi
xi xi

+ k 2 c 2

(2.34)

where the first three terms on the right-hand side are identified as turbulent production,
turbulent diffusion, and turbulent dissipation, respectively. The last term is direct
molecular diffusion of the scalar variance, and is negligible for the small values of
molecular diffusivity relative to atmospheric dispersion scales. However, it is important

12

DRAFT
to note that the only dissipation mechanism is molecular, as is evident from consideration
of the total scalar variance equation

( )

( )

c
c 2

+
ui c 2 =
uic 2 2k
xi
t xi
xi

+ k 2 c2

(2.35)

where it can be seen that all the terms except dissipation are in flux form.

For very high values of the Schmidt number, i.e., very small values of , the
molecular terms in both equations are effectively equal to the turbulent scalar dissipation




c
c = 2k
xi

(2.36)

which is modeled using a dissipation timescale as

c =

c2

(2.37)

From the above, it is clear that the statistical variance in concentration due to
random fluctuations in the velocity field is controlled by the turbulent flux and turbulent
dissipation rate. SCIPUFF employs closure models for these terms to provide a
prediction of the scalar fluctuation variance. The critical aspect of the prediction lies in
the determination of the dissipation timescale, c , associated with the turbulent eddies
responsible for the dispersion, and the recognition that large scale eddies can simply
meander the complete concentration field without producing a reduction in the variance.

2.2.2 Scalar Fluctuation Dissipation Timescale.

The key aspect of the scalar variance prediction is the dissipation timescale.
Proper characterization of the dissipation timescale allows an accurate prediction of the
concentration fluctuation variance, as has been demonstrated in comparison with
laboratory data (Sykes, Lewellen, and Parker, 1986) and also with large-scale
atmospheric dispersion observations (Sykes et al., 1993c). The second-order closure
model for the dissipation rate was originally developed by Sykes et al. (1984) using the
laboratory data of Fackrell and Robins (1982). A fundamental discovery in that study
was that the scalar dissipation scales were an internal property of the scalar field itself,
rather than being determined exclusively by the velocity fluctuations. Essentially,

13

DRAFT
velocity eddies with scales larger than the scalar plume or cloud will meander the entire
scalar distribution but not cause any nonlinear cascade of scalar fluctuation variance onto
smaller scales. The turbulent cascade process is only driven by eddies with scales similar
to the instantaneous plume size. This is not the case for persistent mean wind shear,
which can distort a cloud by means of the continued stretching, and this phenomenon is
discussed separately in Section 2.2.3.
The general form for the scalar dissipation time scale is
1

qc
c

(2.38)

where q c and c represent the velocity and length scales for the scalar fluctuations. The
velocity scale is obtained from a simplified description of the turbulent kinetic energy
spectrum, and the length scale is predicted from a transport equation. Before discussing
the details of these parameterizations, however, we introduce a generalization of the
dissipation model to account for the anisotropy between horizontal and vertical directions,
and the differences between plume and puff meandering.
The scalar variance dissipation model introduced by Sykes et al. (1984, 1986)
uses a single estimate of the velocity and length scales to define the dissipation timescale
in (2.38). This is appropriate for the case of plume dispersion in two dimensions with
near-isotropic turbulence. Under the more general conditions of larger scale horizontal
dispersion and arbitrary sources, we require a characterization of the different dissipation
rates associated with each direction. The vertical direction must be distinguished, and we
also need two horizontal scales. The horizontal information is needed to distinguish
between plume-type sources, where the diffusion occurs in two spatial dimensions, and
puff-type sources, which diffuse in all three dimensions. As part of the concentration
fluctuation prediction, we therefore require three length scales in addition to the
fluctuation variance. The internal scales are denoted as c , cH , and cV for the two
horizontal and vertical directions, respectively.
The initial conditions for the internal length scales will be described as part of the
source definition in Section 7. In this section, we describe the dissipation length and time
scale parameterization. The two scales, c and cV , are used to characterize the
horizontal and vertical concentration fluctuation length scales. The second horizontal
scale is used to determine the dissipation timescale only. We use these two concentration

14

DRAFT
length scales to estimate the dissipation velocity scales, using a simplified Kolmogorov
spectrum assumption to determine the appropriate energy for each of the ambient
turbulence populations. Thus, the horizontal velocity scale for the concentration
fluctuations is modeled as
2
2
2
2
qc = q cL + q cB + q cS

(2.39)

where

2
cL

2
cB

FF
= e u + v j minG G
GH H
FF
= e u + v j minG G
GH H
2

2
B

2
B

IJ , FG IJ IJ
K H K JK
IJ , FG IJ IJ
K H K JK
FF I F
+ w j minG G
GH H JK , GH
2

L
c

(2.40)

(2.41)

2
q cS = u S 2 + v S 2

IJ IJ
K JK
2

(2.42)

and the vertical dissipation fluctuation velocity scale is defined as the smaller of the two
estimates

F F I I
min G
GH , GH JK JJK
1

(1)
cV

= qV

(2
qcV) =

cV

cV

VB

VB

K zeq
bs cV

(2.43)

(2.44)

where the second estimate accounts for the reduced dissipation under stable conditions.
The turbulence model constants, b and s, were introduced in the previous section and take
the values 0.125 and 1.8, respectively (Lewellen, 1977). The equilibrium vertical
diffusivity, K zeq , is defined in (2.28).
Using the two velocity scales defined above, the scalar dissipation timescale is
then modeled as
1

= bs

FG q

qc
q
+ cV
cH cV

IJ
K

15

(2.45)

DRAFT
The internal fluctuation scales grow with the turbulent diffusion, and the internal
velocity scale gives a measure of the growth rate. The model for the two horizontal
internal length scales is

d c d cH
2
2
2
2
2
2
=
= cL q cL + cB qcB + cS qcS
dt
dt

(2.46)

where cL = 0.25 for c < L , otherwise cL = 0.098, and similarly for the other two
components.
The vertical scale growth is represented by

R
|

(1
0.25q cV)
d cV
=
(2
dt
bsq cV)

S|T

(2
(1
, q cV) q cV)
(2
(1
, q cV) < q cV)

(2.47)

2.2.3 Vertical Wind Shear Effects.

The effect of a sustained vertical wind shear is to elongate a cloud along the
direction of the shear, producing a thin 'sheet' of material if the vertical diffusivity is
relatively small. This process is often observed in dispersion in the free atmosphere
above the turbulent boundary layer, and also near the surface under stable (usually
nocturnal) conditions. If the vertical diffusivity is small, then the internal scale of the
cloud can be reduced as the cloud is thinned by the shear, and the effect of the vertical
diffusivity is enhanced. A representation of this effect is included in the SCIPUFF model
equations, based on an analytic result for the shear-modified diffusion problem.
We consider an initial spherical Gaussian of material, with spread 0 , in a wind
shear, , and a constant vertical diffusivity, Kz. We non-dimensionalize time using the
shear rate, so that = t is the time variable, and use 0 as the unit length, so that
ij = 2 ij . Then the dimensionless solution for the concentration field can be
0
expressed in terms of a single parameter

B=

Kz

(2.48)

2
0

16

DRAFT
For a constant shear and vertical diffusivity and zero horizontal diffusion, the
non-dimensional solution to the Gaussian moment equations (2.11) is

11 = 1 + 2 +

2
B 3
3

13 = + B 2

(2.49)

33 = 1 + 2 B

A simplified approximation for the dissipation timescale is obtained from the

logarithmic rate of change of c 2 . This is strictly only valid for a high fluctuation
intensity, since we assume that the total mean square concentration is approximately
equal to the variance, i.e., the square of the mean is negligible. This is the regime where
the dissipation is most critical, however. Using the solution (2.49), the dissipation
timescale is found to be
1

c2

d 2
c
dt

2
B 3
3
2 3 1 4
+ 2 + + B
3
3
1+ 2 +

B 1

(2.50)

The behavior of the dissipation timescale (2.50) is illustrated by the solid lines in
Figure 2-1 for a range of values of B. For small values of B, that is for small diffusivity,
the timescale is almost constant until 1. Since time was non-dimensionalized using
the shear rate, this implies that the distortion must proceed for about one shear timescale
before there is a significant change in the dissipation rate. As the cloud becomes sheared
the timescale is reduced, corresponding to the reduced local thickness, and reaches a
minimum value. The minimum depends on the diffusivity, and smaller values of
diffusivity give a lower minimum (relative to the initial value) at later time. Eventually,
however, the diffusivity dominates as the cloud becomes thinner, and thereafter the
dissipation timescale increases linearly with time.
The dashed lines in Figure 2-1 show the predictions of a simplified model for
cV . Using (2.44) for the velocity scale, we modify the vertical scale equation (2.47) to
include a term describing the shear-driven reduction, giving
d cV
K
3 2cV
= z
cV
dt
2 cH

(2.51)

17

DRAFT

B = 10-5
10-4

10-3
10-2
10-1
1

Figure 2-1. Non-dimensional fluctuation dissipation timescale as a function of

dimensionless time for a shear-distorted Gaussian puff. The solid
line is the analytic solution (2.50). Dashed line is the model
prediction from (2.51) and (2.52).

The shear term depends on the horizontal scale, cH , which is generally increased by the
wind shear, and is modeled as

LM

d cH
K
2cV
3
= z +
cV max 0 , 1 cV
cH K z
cV
dt
2

MN

e j

OP

(2.52)

PQ

In the SCIPUFF model, the terms involving the wind shear, , from (2.51) and
(2.52) are added to the scale equations (2.46) and (2.47) with an appropriate definition of
the shear rate. Note that only one of the horizontal scales, cH , is increased by the
presence of shear. For general meteorology, the shear rate is defined as

18

DRAFT
2

vL2

vL
z

(2.53)

( ) ( )

1 1 2

+
uL
4 uL2 z

v
+
z

( ) ( )

u
=
z
2

This includes an estimate of the large scale fluctuating shear in addition to the mean wind
contribution. The fluctuation contribution assumes a linear velocity distribution with
height as typical of the instantaneous distribution. This is a plausible assumption for large
scale variations, where the thermal wind relation dominates the tropospheric wind profile.

2.2.4 Puff Interaction Treatment.

In addition to the three internal fluctuation length scales discussed in 2.2.3, the
puffs must also include an equation for the concentration fluctuation variance. This is a
nonlinear quantity and therefore requires a treatment of puff interactions. We have found
that it is easier to deal with the total square concentration, c 2 , since the conservation
equation is simpler than the fluctuation variance, c 2 , cf. (2.34) and (2.35). Since the
total square concentration equation only involves a dissipation term, the puff transport
equation is simply
c2
d 2
c =
c
dt

(2.54)

where the dissipation timescale, c , has been defined in the previous section.
Note that the puff integral on the right-hand side of (2.54) is the fluctuation
variance, which must be estimated from the relation
c2 = c2 + c 2

(2.55)

If we represent the mean concentration as a sum of Gaussian puffs, then the

expression for the local concentration is
c x =

Q ( ) G ( ) ( x )

(2.56)

where the sum is taken over all the contributing puffs, denoted by a superscript , and
G ( ) represents the spatial Gaussian distribution of puff- . The square of the mean

concentration is therefore

af

Q ( ) Q ( ) G ( ) ( x )G ( ) ( x )


c2 x =

19

(2.57)

DRAFT
and requires a sum over all puff pairs. The distribution of c 2 amongst the individual puffs
for substitution in (2.55) is not unique, but an obvious partition is given by
( )

= Q ( )

c2

Q ( ) I

(2.58)

where the interaction integral is defined as

I = G ( ) ( x )G ( ) ( x )

(2.59)

i.e., the overlap integral between the two Gaussian puffs.

Ignoring surface reflections for the present, the Gaussian shape is conveniently
specified using the inverse of the spread tensor as
G

( )

exp A ( ) ( x )

( x) =

a 2 f

3/ 2

(2.60)

1/ 2

where D = ( ) and denotes the determinant. The exponential argument is

and

( )
( )
ij

(2.62)

The product of the two Gaussians is then

( x) =

exp A ( ) ( x ) A ( ) ( x )

( x) G

( )

( )

(2.61)

ij =

A ( ) ( x ) = ij x i x i( ) x j x j( )

D D

1/ 2

(2.63)

and the exponential argument can be written in the form

A ( ) ( x ) = A ( ) ( x ) + A ( ) ( x )

= ij x i x i x i( ) x j x j x j( ) + c

where

id

(2.64)

c = 0 ij x i x j

(2.65)

ij = ij + ij

(2.66)



0 = ij i j

(2.67)

x i = ij1 jk k

(2.68)

20

DRAFT
i = x i( ) x i( )

(2.69)

Here ij is the inverse of the moments of puff- , defined similarly to (2.62)

The overlap integral can then be computed directly as

exp ( c )

I =

32
( 4 ) (

(2.70)

D D )

1/ 2

At present, only surface reflections are treated in the puff interactions. As shown
in Section 9.1, the reflected Gaussian distribution for an individual puff is described by a
simple multiplicative exponential factor. The reflected distribution is therefore obtained
by multiplying the original Gaussian by 1 + exp R z , where R is defined in (9.9).
Applying this reflection factor to the two interacting puffs and performing the overlap
integral in the region z > 0 introduces a factor, f R , in the integral I , where

1
erf z + z
2
+e

e d

fR =

c
erf d z

33 + e g erf z 33

erf z 33 + e

33

(2.71)

Here,
z = z + z

z = z + z

z = z + z

R
2 33
R
2 33
R + R
2 33

F
R I
= R G d z+z i
J
4 K
H
R I
F
= R G d z+z i
J
4 K
H
R +R I
F
=d R +R i G d z+z i
JK
4
H

(2.72)

33

33

33

and

33 =

c h

(2.73)

33

21

DRAFT
In principle, the interaction calculation described above should be made for all
possible pairs of puffs. This is clearly impractical for a large number of puffs and is
usually unnecessary since many pairs will be separated in space and the overlap integral
will be negligible. SCIPUFF uses a three-dimensional grid with a multi-level adaptive
technique in the horizontal plane to determine near-neighbor puffs. The technique is
described in Section 3.2 in connection with the merging process. The multi-level
technique allocates each puff to a horizontal grid cell based on the puff size. Interactions
are only calculated for puffs in neighboring cells on this grid. Since puff growth is
limited by the splitting process described in Section 3.1, the size of a puff is constrained
to be less than the grid dimensions. Contributions to the overlap integral from puffs more
remote than the neighbor cells are therefore small.
The overlap integral is only calculated for puffs on the same horizontal grid, so
that the interaction between puffs with different horizontal sizes is neglected. This is
equivalent to an assumption of no correlation between such puffs, since the total variance
is calculated as a sum of the contributions from each puff. If the variance is computed
without interaction, then a simple sum of variances implies uncorrelated fluctuations.
This is a reasonable assumption for puffs with different size, since they must have
different transport histories and are generally uncorrelated.
We also note that SCIPUFF can calculate multiple species and/or particle size
groups in a single computation and each puff is assigned a specific type descriptor to
designate its material properties. Interactions are only computed for puffs of the same
type, and correlations between different types are not calculated. Any subsequent
combination sum of material types or particle size groups must estimate a crosscorrelation if the variance is required. For the special case of multiple particle size bins,
two variance calculations are stored so that statistics for the total material concentrations
are available. SCIPUFF accounts for interactions between all puffs in the same size
group, and also between all puffs of the same material and any size group. This provides
a variance calculation for the concentration of the individual size group, and also a
variance for the total concentration obtained as the sum over all size bins.

22

DRAFT
2.2.5 Conditional Average Dispersion.

The second-order closure framework of SCIPUFF provides a prediction of the

statistical variance in the concentration value, as described above. This variance gives a
measure of the uncertainty in the concentration prediction and depends on the velocity
fluctuation statistics used to determine the dispersion rates. Under some circumstances, a
meaningful quasi-deterministic prediction can be obtained by restricting the definition of
the turbulent velocity fluctuations. By ignoring the large-scale meandering motions in
the turbulence, a conditional average result can be generated in which the uncertainty in
the plume/cloud location is ignored. It must be recognized that the conditional average
simply neglects the meandering component of the turbulence and provides a prediction of
the small-scale diffusion effects.
The conditional average is obtained by specifying an averaging time, Tav. This
time scale is converted into a length scale, av, using the local wind speed, and used to
restrict the velocity variance to the diffusive range of scales. The averaging time should
be thought of as a sampling time, but can also be considered to be the release duration if
this is shorter than the sample time. Thus, an instantaneous measurement or an
instantaneous release both imply a zero averaging time.

A reduction factor is applied to the velocity variances if the length scale of the
velocity fluctuations is larger than the conditional average scale, which is defined as the
larger of av and the instantaneous cloud scale. Turbulent motions on scales smaller than
the instantaneous cloud are always diffusive in character, so this part of the spectrum
must be included in the conditional average. When the instantaneous cloud is smaller
than the conditional average scale, the averaging time determines the range of scales. A
simple power law reduction, consistent with the Kolmogorov inertial range behavior of
the energy spectrum, is assumed. Thus the boundary layer turbulence used in the
dispersion model equations in Section 2.1.3 is defined as



2
B

23

(2.74)


2
BT

v = v

HT
HB

2
B

u = u

HT
HB

2
BT

23

(2.75)

23

DRAFT

VT
VB

(2.76)


w = w
BT
B

23

VT
VB

2
B

w = w
2
BT

43

(2.77)

where the conditional filter scales are



(2.78)

VT = min VB , max cV , av


HT = min HB , max c , av

(2.79)

The new velocity correlations and length scales, defined in (2.74)-(2.79), are then
used in the dispersion model equations. A similar procedure is applied to the large-scale
velocity correlations and length scale.

av = 0.03 V Tav

The conditional average length scale,

time scale as

av,

is defined in terms of the averaging

(2.80)

where V is the local velocity scale, which includes both the mean velocity and the
turbulence. Thus
V 2 = ui ui + ui 2

(2.81)

and the conditional boundary layer component uses the boundary layer turbulence. The
length scale for the large-scale component uses the large-scale velocity variance in (2.81).
The conditional averaging representation is described in detail and compared with
field data by Sykes and Gabruk (1997); the field data comparisons are included in Section
14 of this document.

24

DRAFT
2.3 SURFACE DEPOSITION.
2.3.1 SCIPUFF Implementation.

Surface deposition of the gaseous material introduces a source term into the
concentration equation (2.5) and therefore modifies the puff moment equations. A simple
constant deposition velocity, vD, is used, so the puff mass conservation equation (2.7)
becomes
dQ
= Fs
dt

(2.82)

where the mass flux at the surface is defined as

Fs = v D

c dA

(2.83)

z =0

Other deposition effects for particles and liquid droplets are discussed in Sections 4 and 5.
Integrating the mean surface concentration for the generalized Gaussian puff gives
z =0

Q
exp 33 z 2 + 11 x s2 + 22 y s2 + 2 12 x s y s
2DDH

c dA =

(2.84)

where ij is defined in (2.62) as half the inverse of the puff moment tensor ij , D is the
determinant of ij , and
2
DH = 11 22 12

x s = 13 22 23 12
ys

b
= b

g Dz

(2.85)

23

11 13 12

g Dz

The expression in (2.84) includes a factor of 2 to account for the surface reflection, as
discussed in Section 9.1.
The flux term in (2.82) can be used to define a deposition time scale

s =

Q
,
Fs

(2.86)

which is then used to decay all the other concentration-weighted puff moments, such as
xi uc .
j

25

DRAFT
2.4 SURFACE INTEGRALS.
2.4.1 Surface Dose or Dosage Integration.
The integrated surface dose is one of the optional output choices, and is defined as
t

T x, y, t = c x, y,0, t dt

za

(2.87)

This quantity is accumulated on an adaptive grid as described in Section 3.7.

The probabilistic aspect of the SCIPUFF prediction is also incorporated into the
surface integral fields through a calculation of the variance of the integral, 2 . From
T
(2.87), we obtain the equation for the mean-square fluctuation integral as
t

2 t = dt dt c t c t
T

af

z z
0

a f af

(2.88)

which involves a two-time correlation for the concentration field. We have omitted the
spatial coordinates in (2.88) since the time variation is the only concern in the present
discussion.
We make the assumption that the time correlation is exponential, i.e.,

Tc

t t

c ( t ) c ( t ) = c 2 ( t ) exp

(2.89)

where Tc is the integral time scale for the concentration fluctuations. Substituting in
(2.88), and assuming that the time integration covers the range of the autocorrelation
function, we obtain
t

t = 2 dt c 2 t Tc t
2
T

af

af af

(2.90)

The estimation of the integral timescale for the surface concentration fluctuations,
Tc, is currently based on the simplified meandering plume analyses of Gifford (1959) and
Sykes (1984). These analyses show that the effect of intermittency is to reduce the
integral time scale dramatically from the Eulerian wind fluctuation time scale and also to
introduce a logarithmic correction factor. The intermittency is measured by the
concentration fluctuation intensity, and Sykes (1984) suggests the approximation

26

DRAFT
TcP = TE

c2
c2

ln 1 + 2

GH

c2
c2

(2.91)

JK

for the meandering plume case, where TE is the Eulerian velocity time scale.
For the meandering plume model, the fluctuation intensity can be related to the
concentration fluctuation scale, c , and the scale of the ambient turbulent velocity
fluctuations. The horizontal scale of the turbulence, y , is estimated using an energyweighted average of the three horizontal components. Thus
y

(u
=

2
L

+ vL2 L + u B2 + vB2 B + uS2 + vS2 + wS2 S

2
qT

(2.92)

2
where q T is the total turbulent energy. A generalized approximation for the intensity is

then given by

Ic

F I
=G
H JK
2
y

(2.93)

cH

Using this estimate of the intensity, the time scale (2.91) can be written as

TcP =

TE ln 1 + I c
I c ln 2

(2.94)

a f

for c y . For c > y , we use TcP = TE c y , and the proportionality constants in

(2.94) are chosen to match the two estimates at c = y . We estimate the Eulerian scale
of the turbulence as

TE = 0.7

y
2
u 2 + v 2 + qH

(2.95)

The estimate in (2.94) is appropriate for continuous plumes, but instantaneous

sources may have shorter time scales due to the short duration of passage. We therefore
define a "time-of-passage" time scale as

TcH = 0.7

cH
2
u 2 + v 2 + qH

27

(2.96)

DRAFT
using the second horizontal concentration fluctuation scale. This time scale will be large
for a continuous source, since cH represents the streamwise length scale. The correlation
time scale is defined as the smaller of the two estimates, i.e.,

Tc = min TcP , TcH

(2.97)

The variance contribution is accumulated on the same surface grid as the mean,
using the Gaussian shape of the mean concentration, so that
c 2

T ( t + t ) = T ( t ) + 2c t Tc
2

(2.98)

In addition, the effective length scale of the fluctuations is calculated using the
variance as a weighting factor applied to c .
2.4.2 Surface Deposition Integration.

The mass deposited on the surface is computed in a very similar way to the
surface dose. In this case, the deposition is defined as
t

D x, y, t = v D c x, y, , t dt

(2.99)

where vD is the total deposition velocity, including gravitational settling for particles or
droplets and turbulent dry deposition.
The mean and variance of the deposition are calculated exactly the same as the
dose but with the extra factor of the deposition velocity. Conservation of total mass is
ensured by estimating the deposited mass as the actual loss of mass from the puff in the
time step t.

SECTION 3
NUMERICAL TECHNIQUES

28

DRAFT
3.1 PUFF SPLITTING SCHEME.
The puff moment evolution equations given in Section 2.1 generally increase the
size of the puff through turbulent diffusion and elongation along the direction of the wind
shear. As the puff grows, the local representation of the turbulence and velocity fields
using the puff centroid location becomes increasingly inaccurate.
When the
meteorological fields are inhomogeneous, the accuracy of the calculation can only be
maintained by splitting puffs into smaller components that can sample the variations in
the meteorology explicitly. A grid-based method for splitting puffs was presented by
Sykes and Henn (1992), where the moment method of Egan and Mahoney (1972) was
extended to include some shear effects. In this scheme, grid cells can be thought of as
each containing a Lagrangian puff. As the puffs move and spread into neighboring grid
cells in a time step, the masses are redistributed so as to maintain a single puff within
each cell. We wish to avoid any numerical grid in the current method, however, so the
redistribution cannot be based on the rectangular grid cells.

The three diagonal moments,

, give the effective length of the puff along each

coordinate direction, and we choose to split a puff in the x-direction, say, when
11 > H 2 . Here H represents the limit for the horizontal spread and should be chosen
so that the linearization of the velocity and turbulence fields is valid for a Gaussian with
smaller spread. Similar splitting criteria apply for the other two coordinate directions,
using H in the y-direction and V in the z-direction. We wish to represent the original

Gaussian puff with several smaller, overlapping puffs that conserve all the puff moments
and only change local concentration values by a small amount. For a split in the xdirection, the original puff is replaced by two smaller puffs as follows. The new centroid
locations are displaced by a fraction, r, of the puff spread in the x-direction, and by a
distance proportional to the off-diagonal moment in the other two coordinate directions.
Thus

x1( ) = x1 r 11

(3.1)

(
x 2 ) = x 2 r

12
11

(3.2)

(
x 3 ) = x 3 r

13
11

(3.3)

29

DRAFT

where

= {1, 2} corresponds to the plus and minus sign, respectively. The diagonal

moments for the new puffs are obtained using the following relations

( )
22

(
11 ) = 11 1 r 2

(3.4)

FG 1 r IJ
H K
FG 1 r IJ
H K
2

= 22

11
2

(
33 ) = 33

11

2
12

(3.5)

22

2
13

(3.6)

33

and the new off-diagonal moments are

(23 ) = 23

(3.7)

(
13 ) = 13 1 r 2

(
12 ) = 12 1 r 2

(3.8)

r 2 12 13

(3.9)

11

Conservation of all puff moments is ensured by this procedure, and the moments
of the new puffs are all reduced and are also realizable. The realizability constraints
involve the Schwartz inequality between the diagonal and off-diagonal moments, e.g.,
2
12 11 22 and the strictly positive value for the determinant. These properties can be

verified directly from the above relations.

The generalized puff description contains other information in addition to the

moments discussed above. The additional variables fall into two categories, either a
conserved puff integral property (similar to the puff mass) or a puff value property (such
as turbulence length scale). Integral properties are simply divided equally between the
two new puffs, and the value properties are assigned equally to both.

The reduction in puff size is controlled by the parameter r, but a larger reduction
in puff size reduces the amount of overlap between the new puffs and gives a poorer
representation of the original Gaussian shape. For a puff with diagonal moment tensor,
the splitting scheme described by (3.1)-(3.9) creates a pair of Gaussian puffs with a
dimensionless separation

s =

2r
1 r2

30

(3.10)

DRAFT
1
where is the distance between the new centroid positions and = 11 is the new
spread in the x-direction.

In general, the difference in concentration distribution resulting from the splitting

operation is difficult to specify for an arbitrary Gaussian, but the simple spherical puff
with diagonal moments

provides an idealized case. Figure 3-1 shows the maximum

difference between the original Gaussian and the sum of the two smaller Gaussians after
splitting in one-dimension, i.e., x-direction only, and also for a two-dimensional split in
both x and y, which creates 4 puffs. The difference is given relative to the maximum
concentration value in the original Gaussian, and is plotted as a function of the
dimensionless separation, s. It is clear that the one-dimensional case provides better

overlap between the new puffs and smaller differences for the same separation. A
maximum local concentration change of 20% requires a dimensionless separation of less
than 2.25 for a one-dimensional split, but must be less than 1.8 for two dimensions. In
practice, there will usually be additional overlap from other puffs and we have found that
r = 0.75. i.e., s = 2.25, gives the optimum size reduction with acceptable overlap. In

diffusive applications, the puff size increases after splitting and reduces the effective
separation and overlap errors.

The vertical displacement involved in splitting a puff can cause creation of puffs
below the ground surface, so a reflection condition must be imposed in these cases. The
reflection adjustment is applied if the newly created puff is below the ground. In this
case, the puff centroid is relocated by reflecting the vertical coordinate and positioning
the centroid along the original displacement line, as illustrated in Figure 3-2.

A similar condition is imposed at the boundary layer capping inversion, z = zi, to

prevent diffusion of material through the interface from below. The inversion reflection
is applied when a puff within the PBL splits and creates a new puff with z > z i . In this
case, the puff will be reflected back along the line joining its centroid to the original puff
centroid.

31

DRAFT
0.5

0.4
0.3
2-d split
0.2
0.1
0

1-d split

0.5

1.0

1.5

2.0

2.5

Figure 3-1. Maximum dimensionless error, s, from splitting a single Gaussian puff as a
function of separation distance, s. The error is relative to the maximum
concentration value in the original Gaussian, and the separation distance of
the puffs after splitting is relative to the original Gaussian spread. Results
are shown for both one-dimensional and two-dimensional splits.

Split puff-2 centroid

Original puff centroid

Reflected puff-1
centroid

h
Split puff-1 centroid
Figure 3-2. Schematic illustration of puff reflection for a split below the ground surface.

32

DRAFT
3.2 PUFF MERGING SCHEME.
Unfortunately, the splitting process can rapidly lead to excessive numbers of puffs
unless some form of merging is employed to reduce the numbers. In general, splitting the
puff distribution will create many overlapping puffs that can be merged together as a
single Gaussian. The merging rules are actually much simpler than the splitting rules,
since the only requirement is moment conservation. Thus, if superscripts 1 and 2
represent two puffs deemed sufficiently close to each other to merge, then they can be
combined as follows:

Q = Q (1) + Q ( 2 )

(3.11)

Q x i = Q (1) x i(1) + Q ( 2 ) x i( 2 )

(3.12)

(1)
(2)
Q ij = Q (1) ( ij + xi(1) x j(1) ) + Q (2) ( ij + xi(2) x j(2) )

(3.13)

where xi( ) = xi( ) xi .

Other puff variables are combined according to their type. Integral properties are
combined, like the puff mass in (3.11), as a sum of the two values. Value properties are
mass weighted and combined like the puff centroid, (3.12).

The difficulty associated with merging lies in deciding which puff pairs are
eligible. In general, if we have N puffs, there are N2 possible pair combinations and it
rapidly becomes impractical to consider all such pairs when N is large. Some techniques
have been suggested for ordering randomly located, multi-dimensional data with the
objective of determining nearest neighbors, (e.g., Boris 1986) but we have chosen to use
an adaptive multi-grid location scheme.

The basic concept of the grid location scheme is to assign each puff to a particular
grid box, then a search for near-neighbors can be carried out by searching over the
neighboring grid cells. There are two technical difficulties that arise in the application of
this concept. First, the grid cell size should be determined by the size of the puff itself, so
that a puff overlaps only a finite number of neighboring grid cells. Second, we must
account for an arbitrary number of puffs within each grid cell. Thus, we cannot simply
allocate storage for a fixed grid with a fixed number of puffs in each cell.

33

DRAFT
The first problem is solved by means of an adaptive grid in the horizontal plane.
In atmospheric cases, the vertical resolution is restricted by the stable stratification, so a
fixed grid size of V is used. However, the horizontal scale varies widely so an initial

coarse grid is defined with size 0 = H. Each cell of this coarse grid can be individually
refined to create 4 additional cells with size 1 = 0/2. The new cells are simply added to
the end of the list of cells, and a pointer from the 0-cell contains the location of the first
of the 4 refined cells. This procedure can be continued as far as computer storage allows.

This successive refinement procedure has the advantage that all coarser levels of
the grid are available in addition to the most refined level. Each cell of the grid stores
two integer variables, the pointer for the subsequent refinement (if it exists) and the
number identifier of the puff located in that cell. The choice of refinement level for
locating a puff on the adaptive grid is determined by the largest of the horizontal
moments, i.e., the grid size n = 0/2n such that

n < 4 max n 1

(3.14)

2
where max = max(11, 22 ) . The factor of 4 ensures that n 2 max , so non-neighbor
cells are separated by at least 2 max and their neglect in the overlap calculation of Section
2.2.4 is justified.

A schematic illustration of the puff allocation is shown in Figure 3-3, where a

single puff with index number is located in cell number 12 (denoted by C12). The puff
is allocated to the second level of grid refinement, and the grid system generated by this
puff is also shown. The grid cells are indicated by the C-number and the two numbers
stored for each cell are given in parentheses below. The first number is the first grid cell
of a refined block of four cells, and is zero for no refinement. The second number is the
index of the puff contained in this cell, and a zero indicates no puff in the cell. When
both numbers are zero for a cell, we have omitted the number pair in some cases. Thus,
all cells except C12 contain no puffs and show the second number as zero. Cell 4 shows
the refinement to the first level and points to cell 5, which is further refined and points to
cell 9. Clearly, a large number of puffs will produce a much more complicated grid and
index structure, but this simple adaptive refinement provides an efficient location scheme
for locating arbitrary collections of puffs.

34

DRAFT
In locating the puffs on the grid, the index number of the first puff placed in a
particular cell is stored in the grid cell storage location, but if other puffs share the same
cell then their indices must be saved elsewhere. Unlimited numbers of puffs can be
accommodated by using a linked list; each puff carries a pointer to designate the next
puff in the cell. Thus, if puff number j is found to share a grid cell with a previous
occupant, say puff number i, then the list pointer in puff i is set to point to puff j. In
practice, the entire list for the cell must be scanned and the new puff added to the end of
the list. However, since the cells are adapted to the puff size it is highly unlikely that a
single cell will contain more than a few puffs.

1
C8

C7
0
C3
(0,0)

2
C11

C4
(5,0)
C5
(9,0)

C1
(0,0)

C9

C10

C12
(0, )

puff-
C6

C2
(0,0)

Figure 3-3. Schematic illustration of the adaptive multi-grid for locating puffs. Grid cell
numbers are represented as C-n, and two levels of refinement are shown.
See text for a description of the cell storage rules.

35

DRAFT
The use of the adaptive grid technique in conjunction with the linked list provides
a very efficient method for sorting a large number of puffs with arbitrary locations and
moments, giving a compact list of puff pairs to test for possible merging. We note that a
search over the list in each grid cell finds most of the candidate pairs, although puffs
allocated to different cells but lying very close to each other across the boundary will be
missed. In practical terms, it is more efficient to miss some merges rather than scan
multiple cells. The actual merging of a candidate pair is based on the overlap integral of
the two Gaussian functions, which can be written in the form
I=

a2 f

D12

exp A12

(3.15)

where D12 is a combination of the two determinants and the exponential argument depends
on both puff moments and their separation. This integral is computed as part of the
concentration fluctuation variance calculation (see Section 2.2). For two identical,
spherical Gaussians with separation distance, and diagonal moments, 2 , the argument
can be written very simply as A12 = 2 4 2 . A pair is merged if A12 < 2 4, so that m
m
is a measure of the centroid separation relative to the size of the two puffs. We note that
the separation is not a direct indication that the two puffs are nearly coincident, since they
could be of different size or shape. However, the multigrid sorting scheme ensures that
the pair have similar overall size and it is unlikely that two such puffs will arrive at the
same spatial position with very different shear histories. A more reliable merge criterion
can be developed to account for the shape differences as well as the centroid separation,
but we have not found this to be necessary yet.

As discussed in connection with the splitting process, the merging criterion

determines the extent of puff overlap, since a small value for m will maintain more
overlap before merging a pair of puffs. It is important to ensure consistency between the
split and merge criteria, however, in the sense that a newly created split pair should not
satisfy the merge criterion. In practice, we maintain a distinct gap between the two
criteria, and a reasonable value for m is found to be 1.73 in conjunction with the split
criterion, r = 0.75, i.e., s = 2.25.

In addition to the overlap criterion, merging is restricted to puffs of the same

material type and particle size bin, and position relative to the boundary layer cap, zi.
Puffs within the mixed layer, z < zi , are not merged with puffs outside the mixed layer

36

DRAFT
since this could result in artificially diffusing material across the capping inversion. The
puffs are also required to have similar concentration fluctuation scales, c, so that puffs
of significantly different size cannot be merged even if they are spatially coincident.

3.3 PUFF INVERSION CAPPING.

Under daytime mixing conditions, the atmospheric boundary layer is typically
capped by a temperature inversion, which provides a limit on the vertical extent of the
turbulent layer. The depth of the mixed layer is denoted by zi, and the specification of
this inversion height is described in Section 11.2.4. Although there is entrainment across
this layer, the flux through the inversion is relatively small and the vertical diffusivities
are also usually much smaller in the free atmosphere above the boundary layer. The
simplest assumption is therefore to approximate the inversion as an impenetrable surface
and apply a perfect reflection condition for the puff concentration distribution.
The level at which the reflection is applied cannot simply be defined as the local
inversion height, zi, since the inversion can move up or down as the turbulent layer grows
during the daytime or collapses in the evening. If the capping level were reduced, then
the local concentrations would increase as the mass became compressed into a thinner
layer, and this would violate the mass conservation law. We therefore associate a
capping height, zc, with each puff and adjust the capping level in response to the local
boundary layer changes experienced by the puff. A capping height of zero is used to
indicate no reflection.
We first define a capped boundary layer as one in which the surface heat flux is
non-negative and the overlying atmosphere is stable. .i.e., a positive potential temperature
gradient. The specification of the surface heat flux is described in Section 11.2. Upon
release of a puff, zc is initialized as

zc =

Rz
S

T0

if the boundary layer is capped

otherwise

so that zero indicates no cap. At each timestep the puff cap is modified according to the
following rules:

37

DRAFT
Case A: Boundary layer capped, and z z i
This implies the puff centroid lies inside a capped boundary layer, and if the puff
is effectively contained within the layer then z c = z i . Containment is defined by the
condition

z + 2 z zi
If the puff is not contained then it extends beyond the current boundary layer. In this case
a cap must still be defined since the vertical diffusivity is calculated at the puff centroid
within the boundary layer and upward growth of the puff at this rate is inappropriate. If
the puff was previously capped at z c , then the new cap is set to the larger of z c and z i . If
the puff was not capped, then the new cap is defined as

zc = z + 2 z
Case B: All other situations
This includes puffs above the boundary layer or uncapped boundary layers, such
as stable conditions. If the puff is already uncapped then it remains in that state,
otherwise a check is made to test whether the cap should be moved or completely
removed. If the puff is sufficiently below its cap, i.e., z + 2 z z c , then the cap is
removed. Otherwise, the new cap is defined as
(
z c = zc old ) + w + 0.01 w t

The cap is allowed to move upward due to advection by the mean vertical velocity
component, and also grows due to turbulent diffusion by the vertical velocity fluctuations.
The 1% factor on the vertical velocity standard deviation is included to allow the cap to
dissipate slowly under stable nocturnal conditions.

3.4 FINITE-DIFFERENCE SCHEMES.

The ordinary differential equations for the puff moments and properties are
generally solved using a simple, explicit forward time-difference. However, all linear
damping terms are represented implicitly, so that an equation of the form

d
= F
dt

(3.16)

38

DRAFT
is represented numerically as

t + t t

f af

= F t t t + t

af af a

(3.17)

i.e.

f t
f a1t++Ftaf tf
t a

t + t =

(3.18)

The puff mass, turbulence moments, and mean-square concentration integrals can
all be expressed in the general form (3.16).

The puff centroid is advanced using the second-order Adams-Bashforth scheme

x i t + t = x i t +

af

t
3ui t ui t t
2

af

(3.19)

This technique gives accurate trajectory calculations for idealized velocity fields such as
the solid-body rotation flow, and avoids the numerical diffusion produced by the
trajectory divergence errors of a first-order scheme.
The puff second-moment tensor requires special treatment, since it is important to
maintain a realizable tensor with a positive determinant. The effective volume of the
generalized Gaussian described by the six independent moments, ij, is proportional to
the square root of the determinant D = Det( ), and this quantity is used to determine local
concentrations and interaction terms. The volume of a fluid parcel is conserved in an
incompressible flow, and in the absence of diffusion, conservation of the determinant is
implied by (2.11) if the velocity field is solenoidal. It is very important to conserve the
determinant numerically, and we now describe a simple scheme with this property.

The evolution of ij over one timestep is computed using a sequence of steps,

with the intermediate results from step m denoted as (ijm ) . The initial step, with m = 0,

accounts for the diagonal velocity gradients using the relation

(3.20)




u u
+
x x

(n
(0)
= ) exp t

where the Greek subscript implies no summation. Here, t is the numerical timestep and
the superscript n denotes the value from the previous timestep, i.e., after n timesteps. The


39

DRAFT
vanishing velocity divergence ensures conservation of D after this step. The six offdiagonal velocity gradients are then applied sequentially in pairs using the representative
scheme

(3.21)

u (0)
v (0)
u v (0)

22 + t 11 + t 2
12
y
x
y x
u v
1 t 2
y x

u (0)

23
y

(3.23)

(3.24)

u v
1 t 2
y x
v ( 0 )
13
x
u v
1 t 2
y x

0
(23) + t

(3.22)

1
(23) =

(1)
13 =

(0)
13 + t

(1)
12 =

(0)
12 + t

(1)
22 =

(0)
22 + t

v
v (0)
(0)

11
2 12 + t
x
x
u v
1 t 2
y x

(1)
11 =

u
u (0)
(0)

22
2 12 + t
y
y
u v
1 t 2
y x

(0)
11 + t

(3.25)

and the remaining diagonal moment, 33 , is unaffected by these two components of the
velocity gradient. The remaining two off-diagonal velocity gradient pairs are included in
a similar manner by cycling the indices and velocity components, 1 2 3 , and the
advanced time level is finally obtained as (ij3) . This scheme is more complicated than the
single gradient component used in earlier versions of SCIPUFF, but it reduces the errors
arising from sequential application of the shear components and still preserves the value
of the determinant exactly. The relations (3.21)-(3.25) are strictly only first-order accurate
in time, but the conservation of the determinant is a more important property than the
accuracy for long-term stability of the scheme.


40

DRAFT
3.5 ADAPTIVE TIME STEPS.
The description of dispersion over a wide range of spatial scales, from a small
source up to hundreds of kilometers for example, also involves a wide range of time
scales. Small spatial scales are usually associated with short time scales, so close to a
localized source an accurate description of the dispersion process requires a small time
step. As the plume or cloud spreads onto larger scales, however, the time scales will also
increase. The discrete Lagrangian framework of the Gaussian puff model allows an
efficient treatment of the time dependence since puffs can be advanced individually using
an appropriate step for each one. The only interaction between puffs is through the
overlap integrals in the interaction terms, and these are the only terms that require special
consideration when using multiple time steps.

A large time step, tL, is defined for the calculation. This time step must resolve
the meteorological changes and any other time dependence in externally specified
parameters, such as source variations. The large time step is successively halved to
obtain appropriate puff time steps, and the multiple time steps are forced to bring all the
puffs into synchronization at intervals of tL. External parameters are updated only on
the large time step. Within a large time step, however, an individual puff uses its own
locally determined step to integrate for the period tL.

At the beginning of a large time step, the list of puffs is scanned to determine the
smallest step required for this period. This gives the number of small internal steps for
the period, which must be of the form 2M since the local steps are obtained by successive
halving. Although a puff can change its time step during the tL interval, the smallest
time step is not allowed to fall below the initial level of 2M tL, so that the number of
steps is fixed for the period. In practice, this is not restrictive since the time scales are
usually increasing as the puff grows.

The time step for each puff is chosen as the minimum from a list of requirements.
Thus,
t min t c , tu , t w , t K , t shear

41

(3.26)

DRAFT
where the subscripts, c, u, w, K, shear denote the time step limits for dissipation of
concentration fluctuation variance, horizontal advection, vertical advection, vertical
diffusion, and velocity shear, respectively. These time step limits are defined below.
The dissipation time step is determined from the fluctuation variation dissipation
timescale, as

t c =
where

(3.27)

is defined in (2.38).

The horizontal advection restriction is determined by the horizontal flow speed,

tu =

where

(3.28)

u2 +v2

is the local horizontal meteorological grid spacing. Additionally, if the puff is

contributing to a surface integral computation, its horizontal motion is restricted so that it

does not travel more than one standard deviation in one time step. Thus H is not allowed

to exceed

, min

(3.29)

u2 +v2

min

where

11 u 2 + 2 12 u v + 22 v 2

p = max

is the user-specified minimum surface resolution.

The vertical advection time step is determined as

tw =

where

V
wT

(3.30)

is the vertical puff resolution (specified by the user), and wT is the total vertical

velocity defined as the rate of change of z in (2.10), i.e., the sum of the meteorological
component, the gravitational settling, and the turbulent drift.
The diffusive time step limits the vertical spread of a puff, and is given by

t K =

2K
2K p

(3.31)

42

DRAFT

where K is usually the same as

0.8zi inside the boundary layer,

V
K

. However, since the puff vertical spread is limited to

is not allowed to exceed 0.4zi for puffs located within

the boundary layer. A further option for the SCIPUFF calculation is to treat the boundary
layer as a single layer and prevent puff splitting. In this case the V limit is ignored for
puffs within the boundary layer. The puff diffusivity, Kp, is defined as z w c Q .

Finally, the shear time step limit is determined so that the t2 term in (3.21) does
not change the diagonal puff moments by more than 25%. This limit is applied to each
off-diagonal shear component individually. The diagonal shear components collectively
restrict the time step so that no diagonal moment changes by more than a factor of e1/2.

3.6 STATIC PUFFS.

Continuous releases may involve the generation of many puffs with a small
timestep, since the step size is determined by the source dimensions, which are often
relatively small. Puffs will merge as they grow downstream, but the computational time
can become restrictive if the source continues for a long duration. In these cases, it is
reasonable to approximate the early development of the release as a steady-state plume,
since the timescales of the plume are much faster than the changes in the meteorology.
This could be achieved by means of an initial steady state calculation to march the plume
out to some downstream distance, at which point the puffs could be released into the
general flow field with a larger timestep. The difficulty with this approach is that some
method of incorporating the plume phase into the output displays and surface integrals
would need to be developed, so that the initial part of the plume would not be neglected.
In addition, a plume description would need to be stored in association with each
continuous release.
A more integrated approach uses the concept of "static" puffs; these are puffs that
represent the steady-state phase of the plume, and are not advanced every timestep. The
advantage of this approach is that the quasi-steady plume is represented by the same type
of puff as the general unsteady dispersing field, so that inclusion in the output
calculations is facilitated. The concept is illustrated schematically in Figure 3-4.

43

DRAFT
Mean wind

Normal puffs

Static puffs

Actual source
Effective source

Figure 3-4. Schematic illustration of the use of static puffs to represent a continuous
source. The shaded puff is stored as the release description at the end of the
static phase.

The static puffs are generated by a steady plume calculation, described below, and
are removed and regenerated when the meteorology is updated every large timestep, tL.
The static puffs are also deleted from the calculation when the release terminates. The
steady plume calculation is terminated after integrating the plume for 0.5 tL since we are
assuming that the wind field is fixed for the large timestep. The static integration is also
terminated at 10% of the source duration time, if this is smaller than 0.5 tL, since we
cannot consider the plume to be steady on timescales comparable with the release
duration. Finally, the static integration is not allowed to proceed if the puff splitting
criterion is satisfied, or if plumes from different sources interact since neither of these
phenomena are described by the steady state calculation.

A static integration is initiated by creating the first puff exactly as a normal

continuous release; the specification of continuous releases is described in Section 7.1.
This puff is first copied as a static puff, then a normal timestep if taken to produce the
next static puff. At each integration step, a new timestep is calculated so the integration
is as efficient as possible. When the timestep changes, the puff moments must be
adjusted to reflect the change, since the puff now represents a different duration of the
release. If the timestep changes from t1 to t 2 , we use the change ratio

r=

t2
t1

(3.32)

44

DRAFT
to adjust all the mass-weighted integral puff moments as
c

= r c

(3.33)

The puff moments are also adjusted to reflect the relationship between the timestep and
the streamwise length of the plume segment. Thus
(
(
ij2) = ij1) +

2
ui( 2 ) u (j2 ) t2

ui(1) u (j1) t12

(3.34)

where the superscript on the velocity denotes the value from the previous step. This
adjustment ensures that the puff overlap is maintained as the timestep increases.
The nonlinear interactions in the static puff integration are estimated by correcting
the single puff interaction to account for the difference between a fully three-dimensional
Gaussian puff and the two-dimensional Gaussian cylinder. The cylinder approximation is
more representative of a steady-state plume, and the ratio of the overlap integrals for
these two shape functions is

c2
c

cyl

Gauss

2 s
Ls

(3.35)

where s denotes the streamwise direction, Ls is the cylinder length, and

2 =
s

ui u j ij

(3.36)

uk u k

gives the Gaussian moment in the direction of the mean velocity. For low wind speed
situations, where the turbulent diffusion dominates the mean advection, the Gaussian
approximation is more appropriate than the cylinder, so we define an adjustment factor as
2 s
2 s
q2

1 min
fs =
,1
Ls
Ls
u k uk

FG

IJ

FG

IJ

(3.37)

The factor in (3.37) becomes unity when the turbulence dominates and provides a smooth
transition to the cylindrical plume segment factor (3.35). The puff overlap integrals are
therefore computed from the self-interaction term multiplied by the adjustment factor, fs.

45

DRAFT
3.7 ADAPTIVE SURFACE GRIDS.
The wide range of scales needed to describe the dispersion from a small source is
encountered in the surface dosage and deposition integrals described in Section 2.4.
Close to the source, for example, the deposition field may be very narrow but effects may
be required over a very large domain. Using a uniformly spaced receptor grid to
accumulate the integrals makes it difficult to resolve the values close to the source.
Under specialized circumstances, it is possible to use locally refined grids such as an
expanding radial distribution, but this is impractical for multiple sources and complex
flows.

An adaptive grid technique is employed in SCIPUFF to provide enhanced

resolution wherever it is required. A coarse grid is defined to cover the calculation
domain with a mesh of roughly square cells. Grid refinements are then defined with a
factor of 2 decrease in cell size for each horizontal dimension. Grid cells finer than the
top-level coarse grid are created as required during the calculation.

At each puff timestep, the surface integrals are updated if the puff contribution is
significant. Puffs with centroid locations more than 3 z above the local terrain are
neglected in the surface calculation, and the local terrain is defined as the highest
elevation within the horizontal area of the puff. The latter is estimated as

(3.38)

h
h
+ y
x
y

hmax = h ( x , y ) + 3 x

where the slopes are also estimated at the centroid location. Additionally, a minimum
concentration of interest can be specified, and surface contributions will be ignored if the
maximum concentration in the puff falls below the minimum value.
For puffs that contribute to the surface integrals, a resolution scale is determined
from the smaller of the two puff principal axes in the plane parallel to the local surface.
Thus,

2 =
res

0.5

xx + yy +

xx

yy

46

+ 4

(3.39)
2
xy

DRAFT
where the tensor is defined in (2.62) as half the inverse of the spread tensor, . Note
that over sloping terrain, the tensor is rotated into the local coordinate frame. For a
circular puff, the resolution scale is x 2 .

Having determined the resolution scale, the grid level is selected as the largest cell
that is smaller than res. A scan over cells at the appropriate grid level is then made to
calculate the Gaussian puff contribution at each cell out to roughly 6 from the maximum
surface concentration location. If a cell already exists then the contribution is simply
added; otherwise a new cell is created.

47

DRAFT
SECTION 4
PARTICLE MATERIALS
4.1 PARTICLE MATERIAL PROPERTIES.
SCIPUFF allows several classes of material to be specified; this section describes
the modeling of solid particle materials. Particles require a size description and include
gravitational settling effects, in contrast to gaseous materials, which are absorbed at the
surface with a fixed velocity. Each material is described by an 8-character identifier
which is used for output selection in cases with multiple materials. The user-specified
choices of integrated surface output are also characteristics of each material, and are
specified as part of the material properties.
Particulate materials require a material density, p, and a set of size bin ranges
ri : i = 1, N p + 1 which defines Np bins or subgroups. The underlying concept is a
representation of a continuous particle size distribution (PSD) by a discrete set of size
bins. A unique puff "type" is associated with each bin, so that each size group is tracked
separately through the dispersion calculation. Puffs are only allowed to merge with other
puffs of the same type, i.e., same size bin, so that differential settling effects are
represented correctly. The density and size range are used to determine the deposition
effects as described in Sections 4.3 and 4.4.

4.2 PARTICLE GRAVITATIONAL SETTLING.

4.2.1 Particle Fall Velocity.

Particle fall velocities are computed using an equilibrium assumption and an

empirical drag law relationship. The steady state fall speed is obtained from the balance
between gravitational acceleration and the aerodynamic drag force, giving

4
g p rp3 = Fp
3

(4.1)

48

DRAFT
where rp is the equivalent spherical particle radius, p is the particle material density, and
Fp is the drag force.
The drag force is written in terms of a drag coefficient, c D , as
1
2
a c D rp2 v g
2

Fp =

(4.2)

where a is the air density, and v g is the particle fall speed. The drag coefficient is
parameterized as a function of particle Reynolds number (Friedlander, 1977) as

R 24 c 1 + 0.158 Re h
|
= S Re
|T 0.403
23

cD

Re < 1000
Re 1000

(4.3)

where

2 a v g rp

Re =

(4.4)

and the viscosity of air, a , is assumed to be constant at 1.6 10 5 kg m1 s1. The air
density is computed using the local temperature and pressure to account for variations
with altitude.
The above equations can be solved, using iteration for Re<1000, to determine the
fall speed, v g , as a function of air density, particle density, and particle size, i.e.,

v g = Vg a , p , rp

(4.5)

Any realistic calculation of particle dispersion must recognize a particle size

distribution, so that a puff representation must describe a finite range of particle sizes. If
a particle bin is defined for the range r1 rp < r2 , we define the mean particle diameter,
r , as the arithmetic mean of r1 and r2 and estimate the fall speed variation across the bin
using a three-point distribution with an inverse particle size weighting. Thus the mean
fall speed is given as
Vg ( r1 ) Vg ( r ) Vg ( r2 )
1 1 1
+ +
vg =
+
+
r1 r r2
r1
r
r2

(4.6)

and a fall speed variance can also be defined using the same weighting

FG 1 + 1 + 1 IJ c
Hr r rK
1

2
g

2
+ vg =

b g + V a rf + V b r g

Vg2 r1
r1

2
g

2
g

r2

49

(4.7)

DRAFT
The implementation of the gravitational settling process in the puff model using
the mean and the variance of the fall speed is described next.

4.2.2 SCIPUFF Implementation.

Gravitational settling introduces a source term into the concentration equation

(2.5) and therefore modifies the puff moment equations. The source term can be written
as
S=

( vg c )
z

(4.8)

to represent a downward vertical mass flux. Since S does not vanish at the surface, the
addition of gravitational settling implies a loss of material as it is deposited on the ground.
The mass loss can be written as an integral over the surface area, A, giving
Fs = v g

c dA

(4.9)

z =0

and the puff mass conservation equation must be modified to account for the deposition;
this is described in Section 2.3.
The puff centroid equation reflects the settling velocity directly through the
addition of the vertical velocity component, giving
w c
dz
= w x vg +
dt
Q

af

(4.10)

and the effects of the finite range of particle sizes on the vertical spread must also be
considered. The difference in settling velocity of the different particle sizes within a size
bin causes a vertical separation to develop and hence leads to an increase in the puff
moment, 33 . The variation in settling velocity acts in exactly the same way as the
ambient vertical velocity fluctuations, but with a sustained effect since a particle size is
fixed. The fall speed variation gives a linear growth of the vertical spread and is modeled
simply as

d 33
= Terms from (2.11) + 2 g 33
dt

(4.11)

This is not an accurate representation of particle size variability, but provides a simple
estimate of the increase in vertical spread and improves the ability to describe a

50

DRAFT
continuous size distribution. The particle size bins should contain a relatively small
variation in fall speed for an accurate calculation.
The equation for the squared concentration integral (2.54) also includes extra
damping terms to represent the loss of concentration at the surface and also the diffusive
effects of particle size variations. The equation becomes
c2
d 2
c2
c =
c
dt

FG 1 + IJ
H K
g

(4.12)

cV

4.3 DRY DEPOSITION.

4.3.1 Surface Deposition Process.

Dry deposition refers to non-gravitational surface processes that remove material

from the atmosphere, including collection and absorption phenomena. The surface
deposition rate depends on both the contaminant material properties and also on the
character of the surface and the local meteorology. Gaseous deposition mechanisms are
different from particle mechanisms, and different models are employed for the two types
of material. The particle model is complicated by the dependence on particle size and on
the surface roughness or vegetative canopy description. The SCIPUFF representation is
based on the description in Sykes et al. (1993a).
Dry deposition is typically described in terms of the deposition velocity v d , which
is defined by,

v d = Fc / c

(4.13)

where c is the particle concentration and Fc is the total vertical particle flux near the
surface, that is the sum of the turbulent flux and molecular diffusion flux. This
expression is derived from the scalar conservation equation by assuming stationarity and
horizontal homogeneity (Businger, 1986). Note that c, Fc and therefore, vd should
strictly be defined at a specific reference height, typically 1m to 100m above the surface
or vegetative canopy height.

51

DRAFT
The representation in SCIPUFF is simply achieved by adding the dry deposition
velocity, vd , to the gravitational fall speed, v g in the surface removal calculation (2.83).
The remainder of this section describes the parameterization scheme for vd .
The gaseous dry deposition rate depends on chemical properties of the gas, such
as solubility or reactivity, and is currently specified as a constant value for each gaseous
material. Thus, vd is a fixed input value for gases.
The dry deposition model for particles is based on the approach of Slinn (l982),
who postulates that deposition velocity is equal to a collection efficiency multiplying the
momentum deposition velocity,
2
v d = E u* / ur

(4.14)

where u* is the surface friction velocity and u r is the wind speed at the reference height,
z r . Slinn defines the total collection efficiency, E, in terms of the individual collection
mechanism efficiencies as
E = 1 (1 E B )(1 E IN )(1 E IM )

(4.15)

where EB is the efficiency of the viscous sublayer flow within a millimeter of the surface
elements, EIN represents particle interception and EIM represents particle impaction. This
approach is also taken by Davidson, et al. (l982).
A simple logarithmic profile relation is used to determine the reference velocity in
terms of the friction velocity, which is obtained from the local meteorological wind field
as described in Section 11.1.2. However, under free convection conditions, the mean
wind speed is very small and u* vanishes. The turbulent deposition process does not
vanish in this situation, since individual convective eddies will transport momentum (and
particles) to the surface. For the deposition parameterization, we need to define the
average local friction speed, v* , based on the fluctuating surface momentum flux but
independent of the horizontal direction of the flux. Sykes et al. (1993b) have used LargeEddy Simulation to determine a simple expression for this mean friction velocity under
free convection conditions, and find that a good description is given by

ej

v*
z
= 0.46 0
w*
zi

0 .16

(4.16)

52

DRAFT
where w* is the convective velocity scale, z i is the mixed layer depth, and z 0 is the surface
roughness length. These quantities are defined in Section 11.1.
We combine this free-convection limit with the shear-driven friction velocity in
(4.14) to give a final expression for the deposition velocity as

z zd
ln 1 + r
z0

(4.17)

2
v d = E u + v2

where = 0.4 is von Krmns constant, and z d is the displacement height. The next
section describes the specification of the particle deposition efficiency, E. For the rest of
1
2
the discussion, we shall use the effective value of the friction velocity u + v2 2 in place
of u* .

4.3.2 Particle Deposition Efficiencies.

4.3.2.1 Vegetative Canopy Model. In the region very close to a relatively smooth
surface such as water, a flat plate or canopy elements such as leaves or grass, Brownian
motion can be important in transporting particles to the surface (assuming that surface
roughness elements do not protrude through the viscous sublayer). This is particularly
true for very small particles (diameter<0.1m) which, because of their small inertia, are
able to follow small scale velocity fluctuations. Brownian motion results in the diffusion
of particles, i.e., transport from high to low concentration regions. As discussed in
Lewellen (l985), experimental evidence indicates that the deposition due to Brownian
motion can be expressed in terms of the Schmidt number, Sc = v / D , where v is the
kinematic viscosity of air and D is the particle diffusion coefficient. Values of D as a
function of particle radius can be found in Friedlander (1977). The molecular collection
efficiency is modeled as

E B = f D Sc 0.7

(4.18)

where fD is the ratio of the viscous drag to the total canopy drag including pressure
forces. This factor is needed since the particle deposition scales only with the momentum
transport due to viscous forces, and we use a constant value f D = 1 3 to represent the
canopy effects. We follow Lewellen and Sheng (1980) in using a value of 0.7 for the
power law dependence in (4.18); this is close to Slinn's value of 2/3.

53

DRAFT
Interception is essentially filtration in that particles are deposited when they are
within a particle radius of the collecting elements (the characteristic size of the element is
much greater than the particle radius). Particle inertia is ignored so that the particles are
moving with the fluid around the collecting element. Many interception models are
based on the filtering efficiency of cylindrical fibers in potential flow,

E IN =

2 rp

(4.19)

af

where rp is the particle radius and af is the filter fiber radius (Fuchs, l964). However, for
the more complex situation in a canopy, Slinn considers both "small" collectors such as
vegetative hairs and "large" collectors such as grass blades and pine needles and proposes


  

E IN = f

rp

rp + As

+ 1 f

rp

rp + AL

(4.20)

where As is the characteristic radius of "small" collectors, f is the fraction of total

interception by these collectors and AL is the characteristic radius of large collectors.
Slinn (l982) shows that this formulation is in good agreement with Chamberlain's (l967)
wind tunnel measurements of deposition to grass for his chosen parameters. We use
Slinn's expression with f = 0.01, As = 10 m , and AL = 500 m .
Particles are subject to impaction when they are much smaller than the collection
elements but large enough so that their inertia prevents them from following the fluid
flow around the elements. Collection efficiency by impaction is a function of a Stokes
number, which is the ratio of the particle "stop-distance" to a characteristic canopy
element length. Following Slinn, we define the Stokes number

St =

where

u* g
AL

u* (v g / g)
AL

(4.21)

is the particle relaxation time. Slinn gives the impaction efficiency as

E IM =

St 2
1 + St 2

(4.22)

The bulk canopy model used in SCIPUFF involves an assumption about the
velocity and turbulent diffusivity profiles, since the expression (4.14) gives the local
deposition rate as a function of position within the canopy. A simplified representation
for the dry deposition velocity for the entire canopy is needed for implementation in

54

DRAFT
SCIPUFF. If we assume E is small, then the overall efficiency from Slinn's model can be
written in the form
E=

E
1 E +

a f

(4.23)

where E is defined by (4.15) and is a parameter which in general will vary for different
canopies and velocity profiles. This overall collection efficiency is then used in (4.17) to
determine the dry deposition velocity. Slinn gives some data which indicate that the
range of is approximately 0.1 to 0.3 for a wide variety of canopy types. Fits to the data
of Chamberlain (l967) using (4.23) with = 0.16, along with the full Slinn model, are
shown in Sykes et al. (1993a).

For a vegetative canopy with height hc , we specify a displacement height,

z d = hc 2, and a reference height, z r = 2 hc . Note that the canopy height only affects the
reference heights as used in (4.17). The increased deposition for a deeper canopy is
represented by an increased surface roughness and friction velocity.
4.3.2.2 Rough Surface Model. In situations where there is no canopy, e.g., deposition to
water, bare soil or rock, some modifications to the collection efficiency terms are
necessary. Naturally, there is no interception, so EIN = 0. For EB and EIM, we use a
generalization of Lewellen and Sheng's (1980) model for particle deposition to a smooth
flat surface. Thus, we set
EB = 0.8 Sc 0.7

(4.24)

and
E IM =

AIM
1 + AIM

(4.25)

where
AIM = 0.08 St (1 e 0.42 St )

(4.26)

In this case, the Stokes number is given by

St =

2
u* g

(4.27)

Predictions from this model compare favorably with Sehmel's (1973) experimental data on
deposition to a smooth surface.

55

DRAFT
For the rough surface, we specify a displacement height, z d = 0, and a reference
height, z r = 10 m in (4.17).

4.4 PRECIPITATION WASHOUT.

Airborne particles can be collected by precipitation and carried to the surface in
raindrops or snowflakes. This can be an effective means of removal for small particles,
which would not fall significantly due to gravitational settling alone. The microphysics
of precipitation washout can be extremely complex, involving interactions between small
particles in a turbulent flow, in addition to possible electrostatic or chemical interactions.
Detailed description is impractical so a simplified representation is employed in
SCIPUFF.
Precipitation washout introduces an additional source term into the concentration
equation (2.82) and therefore modifies the puff moment equations. Equation (2.82)
becomes
dQ
= Fs + FR
dt

(4.28)

where the mass flux, FR , is defined as

FR =

(4.29)

and R is the washout time scale (the inverse of the scavenging coefficient). The washout
time scale is also used to decay the other concentration-weight puff moments.
The flux of particulate mass carried by the precipitation depends on the size
distributions of both the precipitation and the particles, the fall velocities of both
components, and also the physical form of the precipitation. We shall consider only rain
and snow types of precipitation in the scavenging model presented here. We model the
scavenging process using a collection efficiency to determine the rate of absorption of
particle material in terms of the collision rate, which is simply the product of the number
densities and the relative fall velocity.
Following Seinfeld (1986), we define the scavenging coefficient as

56

DRAFT

1
R

Dr2 Vr N r E dDr

(4.30)

where Dr is the rain/snow particle diameter, Vr is the precipitation fall velocity, N r is the
precipitation size distribution, and E is the collision efficiency. For rain, we use the
gravitational settling velocity from Section 4.2.1 to determine Vr, while snow is assumed
to fall at a fixed speed of 1.1m/s independent of size (Proctor, 1987).
By assuming a narrow bandwidth for Dr , neglecting higher order terms, and using
(4.30) in conjunction with the precipitation rate,

Pr =

Dr3 Vr N r dDr

(4.31)

the scavenging coefficient can be written as

R1 =

3Pr
E D p , Dr
2 Dr

(4.32)

where Dr is an averaged raindrop diameter which we define as

Dr3 N r dDr
Dr =

(4.33)

2
r

D N r dDr
0

Using the Marshall and Palmer (1948) distribution

N r = N 0 e Dr

(4.34)

the averaged raindrop diameter becomes

Dr =

(4.35)

For raindrops, Seinfeld (1986) relates the parameter (in cm1) to the precipitation rate, Pr
(in mm/hr) as

= 41 Pr 0.21

(4.36)

Eliminating from (4.35) and (4.36) gives

Dr = 7.317 104 Pr

0.21

(4.37)

for rain, while Proctor (1987) gives

57

DRAFT
Dr = 5.215 104 Pr

0.25

(4.38)

for snow. The units of Dr are meters in both the above expressions.
Seinfeld (1986) defines the collision efficiency as
E = E1 + E2 + E3

(4.39)

with
(4.40)

32

(4.41)

E3 =

4
1 + 0.4 Re1 2 Sc1 3 + 0.16 Re1 2 Sc1 2
Re Sc

S* =

1.2 + 1 12 ln 1 + Re
.
1 + ln 1 + Re

St S*
E2 =
St S* + 2 3

E1 = 4r 1 + (1 + 2 Re1 2 ) r

(4.42)

where

a f a f
a f

(4.43)

The dimensionless groups used in (4.40)-(4.42) are defined as

Re =

DrVr a
2a

(4.44)

Sc =

a
a

(4.45)

2 Vr Vp


St =

(4.46)

Dr

r = D p Dr

(4.47)

= w a

(4.48)

where a is the air density, a is air viscosity, is the particle diffusivity, is the
particle relaxation time, Dp is the particle diameter, Vp is the particle fall velocity, and w
is the viscosity of water.

58

DRAFT
By substituting Dr for Dr in equations (4.44) - (4.48), we now have a description
of the washout time scale as a function of precipitation rate and particle size. As shown
in Figure 4-1, Seinfelds washout model compares very well with that of Nieto et al.
(1994).

100

102

R1

(s1)
104

Heavy rain

106
Drizzle

108
103

101

10+1

Dp (m)
Figure 4-1. Scavenging coefficient versus particle diameter. Solid lines denote the
Seinfeld (1986) model described above and the dashed lines denote the
model of Nieto et al. (1994) (with a Marshall and Palmer raindrop size
distribution). The precipitation rate ( Pr ) is equal to 0.5 and 25 mmh1 for
drizzle and heavy rain, respectively.

Figure 4-1, however, illustrates a sharp transition region between particle

diameters of 1 to 10m. This transition is due to the second term in the collision
efficiency, equation (4.41), which vanishes as St S* approaches zero from above; for
59

DRAFT
St < S* the efficiency, E2 is negligible. The model of Horn et al. (1988) and the
experimental data of Sparmacher (1993) both indicate that a smoother transition region is
more appropriate. In order to remove the strong variation with particle size and improve
the agreement with observations, we modify the model and set S* to zero in (4.41). With
this modification, the experimental data suggested that the washout rate was
overestimated; therefore, we empirically reduced our scavenging coefficient by applying
a factor of 1/3 to the total collision efficiency, E. As shown in Fig 4-2, results from the
modified SCIPUFF scavenging model compares well with the model predictions of Horn
et al. (1988).

10 0

102

R1

(s1)
104
Moderate Rain

106
Drizzle

108
103

101

10+1

Dp (m)
Figure 4-2. Scavenging coefficient versus particle diameter. Solid lines denote the
model implemented in SCIPUFF and the dashed lines denote the model of
Horn et al. (1988). The precipitation rate ( Pr ) is equal to 0.5 and 10 mmh1
for drizzle and moderate rain, respectively.

60

DRAFT
Although the washout formulation given above is based on a specification of the
precipitation rate, Pr, the input to SCIPUFF is based on a simplified index, which defines
light, moderate, or heavy precipitation in the form of either rain or snow. The effective
precipitation rates associated with the discrete categories are given in Table 4-1.

Table 4-1.

Precipitation rates in mm/hr associated with precipitation types.

Intensity

Rain

Snow

Light

0.5

Moderate

3.5

20

Heavy

25

100

4.5 EFFECTS ON DIFFUSION.

Heavy particles can only respond on a finite timescale to turbulent velocity
fluctuations and are also falling rapidly through the turbulent velocity field. In general,
the effects of the finite timescale due to particle inertia are insignificant in atmospheric
dispersion, since the timescale is relatively short. The response time is related to the
equilibrium settling velocity as

g =

vg

(4.49)

so that a fall velocity of 10ms 1 implies a response time of about 1s. Atmospheric eddy
timescales are generally much longer than one second, except very close to the surface,
and the dominant effect for larger particles is therefore the vertical motion through the
eddies. If the settling velocity is not much larger than the atmospheric eddy component in
the vertical then the response timescale will clearly be very fast and the particles can be
considered to be in equilibrium. We therefore only represent the fall velocity effect in
SCIPUFF.

61

DRAFT
The main effect of a particle falling through a random velocity field is a
decorrelation of the eddy forcing term. The dispersion rate is determined by the velocity
variance and the Lagrangian correlation timescale, so a simple representation is obtained
by modifying the timescale in the turbulent correlation equations. The timescale is
modeled as S q S in (2.21), and similar estimates are used in the other diffusivity
correlations using the appropriate length and velocity scales. For falling particles, the
timescale is modified by a factor
2
qS
2
2
qS + vg

so that the effective diffusivity is also reduced by this factor.

62

DRAFT
SECTION 5
LIQUID MATERIALS
5.1 LIQUID SPECIFICATIONS.
Liquid materials include similar effects to particles, but also involve evaporative
transformation to the vapor phase. Liquid materials consequently require a set of size
bins together with density and vapor pressure information. The liquid density is used to
determine gravitational settling velocity as a function of droplet size, and saturation vapor
pressure is obtained from the Antoine coefficients, A, B, and C as

log10 p = A

af

B
T +C

(5.1)

where p is in mmHg and T is in oC. Surface tension is also required for the liquid material
so that droplet deformation can be estimated. Vapor phase information includes the
molecular weight and surface deposition velocity. Additionally, a surface spread factor
can be specified to describe the effective droplet size after deposition on the ground; this
parameter is used in the secondary evaporation model.
SCIPUFF represents a dispersing liquid-vapor cloud as a collection of puffs with
a different puff type for each droplet size bin together with an extra type for the vapor
phase. Vapor puffs are treated as gaseous materials for all dispersion phenomena, except
that they maintain a total concentration correlation with all the liquid puffs. The actual
liquid puffs carry additional information describing the radius and temperature of the
droplet, since evaporation continuously changes the size and latent heat effects can
modify the droplet temperature. The user-specified size bins are used only to initialize a
size distribution and to merge puffs. Settling velocity and evaporative effects are based
on the calculated size of the droplets in the puff. Thus an evaporating puff will move
through the size bins as it evolves, generating a new vapor puff at each time step.

63

DRAFT
5.2 PRIMARY EVAPORATION.
SCIPUFF bases its primary evaporation algorithm on that of NUSSE4 (Pennsyle,
1992). Several code modifications decrease the computational expense of the algorithm
and adapt the routines to the SCIPUFF framework. These revisions include redefining
the treatment of droplet size distribution, simplifying the calculation of shape
deformation effects, minimizing the computation of air and liquid properties, and
customizing the generation of primary vapor puffs.
Since liquid droplets change their size distribution due to evaporation, the fixed
size bins used to describe particulate materials in the previous section is not applicable.
Each puff in SCIPUFF represents a finite range of droplet size, and this size evolves with
time as determined by the primary evaporation model. The finite size range is used to
represent the vertical spread due to differential fall speeds, analogous to (4.11) for
particulate materials, and the range is held fixed at the initial value, which is determined
by the initial bin size when the puff is released. As for particles, this is only a crude
estimate of the differential fall speed effect.

5.2.1 Droplet Motion and Evaporation.

The SCIPUFF calculation of droplet terminal fall velocities, emulating that of

particles, uses an equilibrium assumption and an empirical drag law relationship.
However, the fluidity of droplet makes necessary the inclusion of shape deformation
effects in the calculation of fall speed. Like particles, a method for calculating the liquid
droplet fall velocity evolves from a balance between gravitational and aerodynamic drag
forces,
4 3
rd d g
3

(5.2)

where Fd is the drag force, rd is the spherical droplet radius,

Fd =

is the liquid droplet

material density, and g is the gravitational acceleration.

The drag force, in terms of a drag coefficient cD, is
Fd =

1
2
a c D re2 v g
2

64

(5.3)

DRAFT

where

is the air density, re is an equivalent droplet radius and vg is the droplet fall

speed. The drag force equation requires an equivalent radius since the force applies to the
projected cross-sectional area of the deformed drop. The parameterization of the drag
coefficient as a function of Reynolds number is the same as (4.3) for particles, using the
equivalent radius in the definition.
The equation for the equivalent radius, re, as a function of Eotvos number, Eo, is
given by
re =

R
|

, Eo 0.4

rd

S| r e 1 + C b E
T
d

0.4

45 13

g j

, Eo > 0.4

(5.4)

where
Eo = 4 g d a rd2 d

(5.5)

is the liquid surface tension, and C1 is chosen to best fit the more complicated NUSSE4
representation. The solution of the above equations, using iteration for Re < 1000,
determines the fall speed, vg.

Evaporation rates depend on the Sherwood number, which is the ratio of the mass
transfer rate and one-half of the diffusion-controlled rate:
m
2Dd rd (Cs C )

Sh =

(5.6)

where Dd is the liquid droplet diffusivity, Cs is the droplet saturation concentration, and
C is the ambient concentration. By definition, this ratio is equal to 2 for diffusioncontrolled evaporation. However, for convection controlled evaporation, the Sherwood
number is greater than 2 and approximated as a function of the Reynolds number and
Schmidt number, Sc, with the expression
Sh = 2 + 0.552 Re Sc 1 3

(5.7)

where the Sc is the ratio of air viscosity to liquid droplet diffusivity, a D d . From
equation (5.6), the mass transfer rate expression is
m = 2 Dd rd (Cs C ) Sh

(5.8)

When shape deformation becomes important, the Sherwood number also becomes
a complicated function of droplet aspect ratio. However, to simplify the calculation,
SCIPUFF uses (5.4) to determine the appropriate Reynolds number for (5.7).
65

DRAFT
According to the Reynolds analogy, which holds for liquids of low volatility, the
heat transfer obeys an equation akin to that of the mass transfer, with the Nusselt number
replacing the Sherwood number and the Schmidt number replacing the Prandtl number.
The Prandtl number is a , where is the thermal diffusivity of air. By applying an
energy balance for the droplet, the droplet temperature, Td, is
Td = T

Sh Dd d
(Cs C )
Nu a

(5.9)

where T is the ambient temperature, d is the latent heat of vaporization of the droplet,
and a is the thermal conductivity of the air.
To further decrease the computational expense of the evaporation algorithm, the
viscosity and thermal conductivity of the ambient air assume constant values of 1.6 105
1 1
1 1
kgm s and 0.0237Wm K , respectively. Additionally, the surface tensions of the
liquid agents are constant at their values at a temperature of 0 oC.

We determined the appropriate value for the empirical constant C1 in (5.4) by

comparing the fall velocity, Sherwood number, and Nusselt number calculated by
SCIPUFF and NUSSE4 for liquid agents over a range of droplet diameters and release
heights. Figure 5-1 shows plots of these parameters for droplets of agent 'VX' at release
heights of 100m and 10km. A value of 0.7 for C1 gave the best fit to the NUSSE4 results.
The time evolutions of the diameter, height above the surface, fall velocity, and
temperature for a range of initial droplet sizes are shown for both SCIPUFF and NUSSE4
in Figure 5-2. The plots demonstrate that the SCIPUFF algorithm for liquid droplet
evaporation reproduces well the results of the NUSSE4 calculation.

66

DRAFT
20
(a)

(b)

16

V 12
(m/s)
8
4
0
120
100

Sh

80
60
40
20
0
70
60
50

Nu

40
30
20
10
0
0.1

D (mm)

10 0.1

10

D (mm)

Figure 5-1. Terminal velocity, Sherwood (Sh), and Nusselt (Nu) numbers for agent VX.
Solid line is SCIPUFF algorithm, short dash lines are from NUSSE4. Left
column (a) is at 100m altitude, column (b) is at 10km.

67

DRAFT
.1 mm

10 mm

101
D
(mm)
10-1

3 mm
1 mm

.3 mm

H
(km)
4

.3 mm
.1 mm

1 mm

10-3

10 mm

3 mm

300
10 mm

16
10 mm

3 mm
1 mm

260
8

3 mm

V
(m/s)

1 mm

T
(K)

.3 mm
.1 mm

1000

2000

3000

220

.3 mm
.1 mm

1000

2000

3000

Time (s)

Time (s)

Figure 5-2. Evolution of droplet diameter (D), altitude (H), fall velocity (V) and droplet
temperature (T) for GB droplets released at a height of 10km with different
initial diameters. Solid line is SCIPUFF, dashed line is NUSSE4.

5.2.2 Primary Vapor Puff Formation.

Primary vapor puffs evolve as the liquid droplet puffs translate and vaporize.
These new gaseous puffs advect and diffuse like other Gaussian puffs, so we must track
them along with evaporating liquid puffs. Hence, the vapor puffs' centroid location,
spread, mass, and concentration variance must be initialized as the puff forms.
The centroid location of a new vapor puff- formed by the evaporation of liquid
puff- is

68

DRAFT

x1 = x1

x 2 = x 2

(5.10)

(5.11)

x3( ) = x3( ) + vg t 2

(5.12)

and the horizontal and vertical spreads are

(5.13)

22 = 22

11 = 11

(5.14)

33( ) = 33( ) + ( vg t 2 )

(5.15)

where vg is the droplet puff fall velocity and t is the evaporation time step. A horizontal

correction to the centroid or spread is not necessary since both the liquid droplet and vapor
puffs move with the mean wind.
The masses of the liquid puff- and vapor puff- after an evaporation time step

are
(5.16)

= 1 r Q


Q = rQ

a f

(5.17)

where the asterisk denotes the state after the evaporation time step t, and r is the ratio of

evaporated mass to original mass. Q is defined for a liquid puff as

Q = cV

Q = cL

(5.18)




and for a vapor puff as



(5.19)




The variables cL and cV represent the liquid and vapor puff concentrations, respectively.
We shall define the total concentration, cT, as the sum of the liquid and vapor
concentrations. The radius of the liquid puff- is reduced by the cube root of r to account
for the mass loss, and the droplet temperature is stored as Td from (5.9).

We next define the effect of evaporation on the fluctuation variances associated

with the liquid and vapor puffs. In order to represent these effects, SCIPUFF stores two
correlation terms for each puff; the first is the correlation with puffs of the same type, i.e.,
liquid-liquid or vapor-vapor, and the second is the correlation with the total liquid plus
vapor. These two correlations are similar to the particle treatment where the single group

69

DRAFT
correlation, c 2 , is calculated in addition to the total concentration correlation, c c T ;
for particles, cT is the sum over all the size groups.
For the liquid puff involved in the evaporation step, we define

2
= cL

(5.20)

= r c L cT

c L cT

2
cL

(5.21)

while the new vapor puff correlations are defined as

c L cT

c L cT

a f FH
= a1 r f
= 1 r

cV c T

2
cV

2
cL

(5.22)

(5.23)

The evaporation process does not change total concentration, so the total correlation is
simply partitioned between the two puffs according to their mass ratio. The liquid
correlation is left unchanged at this point, since the final value can only be determined
after the overlap calculation when the effects of other evaporating puffs is accounted for.
The vapor correlation is initially set to the mass fraction times the original liquid-vapor
correlation, but this also requires adjustment after the overlap calculation step.
The puff interaction calculation provides overlap integrals for the mean
concentrations, as required for advancing the turbulent correlations. For a liquid puff, we
estimate the square of the mean liquid concentration as
=Q

liquid

(5.24)

where the sum is taken only over liquid puffs, and I

is the overlap integral between

and liquid puff- . The interaction integral coefficients are described in

liquid puff-



cL 2

Section 2.2.4. The total mean overlap is similarly given by




&

liquid

I +

&




!"




=Q

Q I

vapor

%$




c L cT

(5.25)

The mean square liquid concentration correlation is adjusted on the basis of the
change in the square of the mean. We calculate the overlap using both the new liquid
puff masses and the old values, which are stored with the puff. The difference between

70

DRAFT
the two is solely due to evaporation. The liquid-liquid correlation is therefore multiplied
by the factor

cL 2

cL 2

(5.26)

old

where the subscript "old" denotes the value computed using the pre-evaporation masses.
For an isolated liquid puff, which only interacts with itself, the factor is easily seen to be
r2, where r is the mass reduction fraction from the evaporation step.
Vapor puffs are adjusted in a similar manner, but is slightly complicated by the
fact that only the liquid puffs retain the "old" mass since only liquid puffs evaporate. The
vapor-vapor correlation adjustment is therefore based on the vapor-liquid overlap
integral. We define

adjusted

= cV c L

cV c L

(5.27)

where

cV c L

cV c L

(5.28)

old

and the denominator is computed using the "old" liquid puff masses. The vapor-liquid
correlations are easily obtained from the two stored correlations, using the relation
= cV c T

cV 2

cV c L

(5.29)

5.3 SECONDARY EVAPORATION.

After a liquid droplet puff impacts the surface, the volatile agent continues to
evaporate by a combination of diffusion and convection. The calculation of this surface
(or secondary) evaporation and the associated vapor puff formation requires additional
surface grid variables, namely the mass-mean droplet radius and wetted area. We
visualize surface evaporation of consisting of three distinct processes: surface grid
deposition, surface grid evaporation, and vapor puff formation.

71

DRAFT
5.3.1 Surface Grid Deposition.
First, we consider the deposition of a single liquid puff. The calculation of the
mass, variance, and scale deposited on the surface is identical to the method described in
section 2.3; however, we use a slightly different notation in this section. After time step
t , the mass (per unit area) deposited at location (xi,yj) by liquid puff- is

d ij = Vd cij t

(5.30)

where Vd is the deposition velocity, cij is the local concentration, and the superscripts
denote that the contributions are from puff- . Therefore, the accumulated mass after t is

d ij = dij + d ij

(5.31)

The surface evaporation model also requires a mass-mean radius and wetted-area
on the surface grid. The mass-weighted radius after t is

Rij = Rij + d ij rs

(5.32)

where rs is the droplet radius as it rests on the surface and is calculated as

rs = s f rd

(5.33)

In (5.33), sf is the liquid agent spread factor and rd is the spherical droplet radius before it
impinges on the surface. The spread factor must be experimentally determined for a given
agent and surface type, and is specified as one of the liquid material properties. Realizing
that the mass of a single droplet is

d i

(5.34)

 

m d = 4 3 d rd

we calculate the droplet number density as the ratio of puff- deposition mass, dij , and
droplet mass, md. It follows that the total wetted surface area accumulates as



  

Wij = Wij + d ij

Ad
md

(5.35)

where the contact area, Ad, is

 

 

Ad = rs

d i

(5.36)

72

DRAFT
5.3.2 Surface Grid Evaporation.

After a liquid agent puff deposits on a surface cell at location (xi, yj), secondary
evaporation begins. Although evaporation occurs on each cell that contains liquid
droplets, creating vapor puffs from all grid cells is computationally expensive.
Therefore, we seek to limit the number of vapor puffs created from surface evaporation,
yet maintain an accurate representation of the process. This is accomplished by scanning
through the adaptive surface grid cells and only releasing puffs at grid levels that are at
least two levels coarser than the finest resolution at the given surface location. The vapor
release is integrated over the 16 cells that are two levels finer than the release level. The
deposition mass, deposition variance, mass-weighted radius, and mass-weighted wetted
area on the release cell are defined as

Vij

1
16

(5.39)

Wij

(5.40)

1
16

(5.38)

Rij

1
16

(5.37)

W =

d ij
i

R=

V =

1
16

d =

where the overbar denotes an average over the sixteen subcells.

After evaporation from the deposition grid, the accumulated mass decreases as

d ij = d ij 1

m
d

(5.41)

where m is the mass per unit area evaporated from the surface. Similarly, the deposition
variance, mass-weighted radius, and mass-weighted wetted area also decrease with
evaporation, i.e.,

m
d

m
d

Rij = Rij 1

Wij= Wij 1

m
Vij = Vij 1
d

(5.42)

(5.43)

K
I

(5.44)

K
73

DRAFT
so that the effective droplet radius and wetted area remain constant during the evaporation
process. In these equations, the averaged total evaporated mass per unit area is
m = m

W
t
R 2

(5.45)

and m is the mass evaporated from a single droplet.

A literature search of droplet evaporation models (to calculate m ) produced only

a limited number of algorithms. The model used in SCIPUFF was suggested by Baines
and James (1994) for evaporation of a droplet from an impermeable surface. However,
the structure of the surface evaporation algorithm allows for new models to be easily

substituted when they become available.

The Sherwood number is the basis of the Baines and James model. For the
present problem, the Sherwood number has the form

m
4 Drs ( Cs C )

Sh =

(5.46)

where D is the diffusivity of the vapor in the surrounding air, Cs is the droplet saturation
concentration, and C is the ambient vapor concentration. Empirical evidence suggests
that

Sh = f , Re, Sc

(5.47)

where Re is the Reynolds number, Sc is the Schmidt number, and is the droplet contact
angle. We combine the Re and Sc numbers to form the Peclet number so that
Sh = f , Pe

(5.48)

Pe = 4 S rs2 D

(5.49)

a f

with

In this equation, the shear rate in the viscous sublayer near the surface, S, is
S=

2
u

(5.50)

and u is the friction velocity.

Furthermore, if we assume that the Peclet number is large, the Sherwood number
approximates as a function of Peclet number only. Baines and James (1994) use

74

DRAFT
boundary layer techniques of the two-dimensional planar problem to approximate the
three-dimensional solution. They found that the assumption of constant radius during the
evaporation process is as acceptable as constant contact angle, and is therefore used in
(5.43) since this is the simplest approximation. Using the constant radius approximation,
Baines and James show that for Peclet numbers much greater than 1, the Sherwood
number simplifies to
Sh = 0.105 Pe 1 3

(5.51)

5.3.3 Vapor Puff Formation.

Secondary vapor puffs form at the surface as the deposited liquid evaporates
through diffusion and convection. Once these puffs develop, they translate and diffuse
passively like other vapor puffs. Like their primary vapor puff counterparts, their
centroid location, size, mass, and concentration variance must be initialized.
The centroid location of a new vapor puff- formed by secondary evaporation
from a surface cell centered at (xi, yj) is

1
x1 = x i + u s t
2
1
x 2 = y j + v s t
2

(5.52)

(5.53)

x 3 = h

(5.54)

where us and vs are the horizontal surface velocities, and h is the local terrain elevation.
The horizontal spread is
(
xx ) = x 2

(5.55)

(yy ) = y 2

(5.56)

where x and y are the cell dimensions. The vertical spread is obtained by assuming that

it is only a function of vertical diffusion, i.e.,

zz = u t A

(5.57)

where is von Krmns constant and A is the turbulence closure constant, with a value of
0.75. Off-diagonal moments are initially set to zero, but in the presence of non-uniform

75

DRAFT
terrain, the newly created puff must be rotated into the local surface coordinates before
release. This ensures that the initial puff is aligned parallel to the surface with a normal
spread defined by the local diffusivity normal to the surface.
We compute the mean-square concentration variance of a puff created by surface
evaporation, with contributions from the variance of the deposited mass and other
interacting puffs, as

1+

Vij
d ij2

= c2

where c 2

c2

(5.58)

is the square of the mean concentration and is described in section 2.2.4.

76

DRAFT
SECTION 6
DYNAMIC EFFECTS
6.1 BUOYANCY AND MOMENTUM RISE.
6.1.1 Mean Flow Dynamics.
Many releases into the atmosphere involve a vertical rise phase since they are
associated with heat or vertical momentum. This includes vertical jets and buoyant
plumes from continuous sources, and buoyant clouds from explosive releases. The
buoyant rise is important since it can strongly influence the effective release height and
the initial source size for the subsequent dispersion in the atmosphere. While the rise
dynamics can be incorporated into the source specification for a passive dispersion
calculation, there are several reasons for providing an internal representation in the puff
dispersion code.
First, the duration of the rise phase may preclude treatment as a source term, since
this phase is ignored by the dispersion calculation. Initialization at the end of the
dynamic rise may be some considerable distance downstream of the source, and effects
calculations in the intervening region would need to account for the rise phase. Second,
the transition to a passive dispersion code requires an arbitrary choice for the position at
which the dynamics can be neglected; an internal representation allows a continuous
decay of the dynamic effects. Furthermore, the dispersion calculation provides a more
consistent treatment of the meteorology, which can be inhomogeneous in both space and
time. Finally, simplified algorithms for plume and puff rise may not be appropriate for
finite duration or time-dependent sources.
In order to represent the buoyant rise dynamics, additional variables must be
stored for each puff. The dynamic model is based on the conservation equations for both
buoyancy, p , and vertical momentum, w p , where is the potential temperature, w is

the vertical velocity, and the subscript p denotes the dynamic puff perturbation from the
local ambient atmospheric value. The angle brackets denote the volume integral over an
individual puff. Only gaseous materials (including the vapor phase of a liquid material)

77

DRAFT
can carry mean dynamics, since we envision the materials being transported by a fluid
medium. As we shall show below, the dynamic gas puffs will transport particles or any
other material if the puffs overlap spatially.
The evolution of the mean dynamic variables is based on the Boussinesq
momentum equation and the conservation of potential temperature, giving
d
g
p
wp =
dt
T0

(6.1)

d
p = B wp
dt
z

(6.2)

where B is the ambient potential temperature. These equations neglect the pressure
gradient in the momentum equation, and simply represent the buoyant acceleration term in
(6.1) and the change in buoyancy due to vertical motion through a potential temperature
gradient in (6.2).
The mean dynamic variables in (6.1) and (6.2) determine the equilibrium rise of a
buoyant puff in a stable atmosphere, where the ambient temperature gradient is positive.
The solution to (6.1) will oscillate indefinitely about the equilibrium level. The vertical
velocity itself will be damped, due to turbulent entrainment, which increases the effective
volume of the integrals in (6.1), but the integrated momentum and buoyancy will be
undamped. While the buoyancy is strictly conserved, the integrated vertical momentum
will actually be reduced due to the generation of gravity waves, and the oscillation will be
damped. We represent this effect with a simple linear damping on the vertical velocity
integral, so that (6.1) becomes
d
g
p cw N w p
wp =
dt
T0

(6.3)

where N is the Brunt-Vaisala frequency, and cw = 0.1. N is only defined non-zero for
stable temperature gradients, and the damping coefficient is arbitrarily chosen as 0.1 to
provide a reasonably rapid adjustment to the equilibrium height.
Tracking the independent evolution of the dynamics of each puff is not sufficient,
however, to model the buoyant rise effects since these effects are interactive. A plume of
buoyant gas rises differently from an isolated buoyant puff because the flow field induced
by the buoyancy has an influence on other parts of the plume. These effects are

78

DRAFT
represented by means of correlation integrals for the momentum and buoyancy, which
account for turbulent fluctuations in addition to the nonlinear interactions. The
correlation equations are written as
wpc

d
g
p c cw N w p c
w pc =
c
dt
T0

(6.4)

p c

d
p c = B wp c
c
dt
z

(6.5)

where c is the fluctuation dissipation timescale, defined in (2.40), which must be

modified to account for the dynamically-induced turbulence. The turbulence model will
be described in the next section.
Note that the turbulent dissipation terms, the last terms in equations (6.4) and
(6.5), involve the fluctuation correlations, which are defined as
w c = w pc w pc
p

(6.6)

p c = p c p c

(6.7)

The mean overlap integrals, w p c and p c , are calculated in SCIPUFF in a similar

manner to the mean concentration overlap terms, c 2 , as described in Section 2.2.4. This
effectively assumes the same Gaussian distribution for both vertical velocity and potential
temperature. The reflection assumption on the vertical velocity field at the surface is a
zero value condition, however, rather than a zero slope.
As noted above, only gaseous materials can carry mean vertical momentum and
perturbation temperature integrals. However, all puffs carry the dynamic correlation
integrals in (6.4)-(6.5) since their species concentrations can be correlated with dynamic
variables due to overlapping with gaseous puffs. This phenomenon is accounted for
through the puff interaction calculations, where the overlap integral with dynamic
gaseous puffs is computed for every type of puff. Thus, passive gas materials or particle
materials co-located with a buoyant gas source will experience the same vertical velocity
field as the buoyant gas itself.
The dynamic correlations provide an additional vertical velocity component,
which is added to the ambient velocity field used to transport the puff centroid. Thus, the
equation for the vertical component of the centroid location becomes

79

DRAFT
w ic
dz
=w x +
+ wp
dt
Q

af

(6.8)

where Q = c is the integrated species mass, w is the ambient vertical velocity, and the
dynamic rise velocity is given by
wp =

wpc

(6.9)

We note that the model only represents the vertical component of the internal puff
momentum, so that the rise can be calculated. The horizontal velocity is not perturbed
from the ambient wind field.

6.1.2 Buoyant Gas Representation.

Buoyancy effects can arise directly from a density difference between the ambient
air and the gas material. This is particularly important for dense gas effects, discussed in
the next section, but is introduced here for the simpler case of buoyant rise. If the gas is
lighter than air, then it will tend to rise without any temperature perturbation. In this
case, the buoyancy variable is actually

bp = g

FG m IJ
0

HT

(6.10)

where m is the mass mixing ratio, 0 is the ambient air density, and is the density
difference gas 0 . Since the local species concentration value, cgas, is equal to m gas ,
we can define a buoyancy factor
Bgas =

(6.11)

gas 0

and write the buoyancy term as

FG B

bp = g

gas gas

HT

IJ

(6.12)

In order to treat non-Boussinesq density perturbations associated with very dense

gases, we modify the gravity term to give

T0

Bgas c gas

1 + Bgas c gas

80

bp = g

(6.13)

DRAFT
so that the effective acceleration due to the gas density difference cannot exceed g. The
correction factor in the denominator is only applied for positive density perturbations,
since lighter-than-air materials must still accelerate the surrounding air mass
For cases with buoyant gas materials, i.e., Bgas 0 for at least one gas material,
every puff must calculate an overlap integral with all the buoyant gases. Thus

bp c = c

(6.14)

Bgas c gas
gas

where the sum is taken over all overlapping gas puffs.

equations are then modified to give

The dynamic puff integral

gB c
d
g
cw N w p
wp =
p + bas
dt
T0
1+ B

(6.15)

for the buoyant gas puffs, and

w c
g bpc
d
g
p
wpc =
cw N w p c
pc +
dt
T0
c
1+ B

(6.16)

bpc

d
bpc =
dt
c

(6.17)

for all puffs. The non-Boussinesq density correction is defined as

B = max 0,

GH

bp c
c

I
JK

(6.18)

Note that the buoyancy correlation term is simply damped toward the mean overlap value,
which is computed in the same way as the temperature overlap.

6.1.3 Turbulent Entrainment.

The velocity field induced by the dynamic rise of a cloud is turbulent, and the
ambient diffusion is enhanced by the internal turbulence of the cloud. The additional
entrainment and dissipation is modeled using an estimate of the internal turbulent
velocity and length scales. The entrainment model is based on earlier work on power
plant plume rise (Sykes et al., 1988) and relates the turbulent velocity to the vertical rise
rate. We define

81

DRAFT

where V is the ambient wind speed, and

q 2 = Fp Ri p w 2 c q1 + c q 2
p
p

V2
V 2 + w2
p

Fp = 1 + 4 Ri p

(6.19)

(6.20)

is a buoyancy factor, representing the increased turbulent energy production due to

gravitational instabilities. The puff Richardson number is defined as
Ri p =

g p p T0 bp
2
wp

(6.21)

where p is the length scale of the internal turbulence. The dynamic temperature and
buoyancy variables are defined similarly to the dynamic vertical velocity as

p =

bp =

pc

(6.22)

Q
bpc

(6.23)

The dynamic length scale, p, is defined equal to the length scale, c from
Section 2.2.2, since this represents an instantaneous scale for the dispersing cloud. The
empirical coefficients in (6.19) are chosen as cq1 = 0.4 and cq 2 = 3.0 from fitting jet rise
data, as discussed in Section 14.3. Additionally, an upper limit of 2 is imposed on Fp to
prevent excessive entrainment velocities.
The dynamic velocity and length scales are used to define a dynamic diffusivity
that is added to the spatial moment equations (2.11) to give
d ij
dt

= ik

u j
xk

+ jk

xiu j c
x j uic
ui
+
+
+ 2 Kp
Q
Q
xk

(6.24)

where
K p = 0.15 q p p

(6.25)

The internal dynamics also provide additional turbulent dissipation, and we

modify the dissipation rate as

82

DRAFT

qp q p
qp
qc
q
q
+ c + cV + 2 bs
+
+
c cH cV
p c cH

(6.26)

= bs

The internal turbulence also modifies the rate equations for the internal length scales, so
that we obtain
d c d cH
=
= cH qc + 0.15 q p
dt
dt

(6.27)

where the first term arises from the ambient turbulence (2.41). The same additional
velocity is added to vertical internal scale equation.

6.2 DENSE GAS EFFECTS

6.2.1 Mean Flow Dynamics.

The puff treatment of buoyancy-driven dynamics described in the previous

section must be modified for the description of dense gas dispersion. The most important
feature of dense gas dynamics is the interaction with the solid ground surface, which
causes lateral spreading as a dense cloud collapses and suppresses the vertical diffusion
due to the stable buoyancy distribution. We first address the dynamic spreading effect,
where the gravitational forcing term drives motion parallel to the ground instead of a
vertical acceleration. The gravitational force is still directed vertically, but the lateral
motion is induced by the pressure gradient, which was ignored in the buoyant rise
situation. We therefore require a treatment of the equations of motion that include the
surface-induced pressure gradient.
A convenient starting point for modeling the dynamics is the vorticity equation,
obtained by taking the curl of the momentum equation,
g j
i
i
ui
2 i
+ uj
=j
+ ijk
+
t
xj
xj
T0 xk
xj xj

(6.28)

where = u is the vorticity and is the kinematic viscosity. The buoyancy forcing is
written using the Boussinesq approximation in the above equation, and we note that the
pressure gradient term vanishes. Forming the vertical component of the vorticity moment

P = e3 x

(6.29)

83

DRAFT
where e3 is the unit vector in the vertical direction, and integrating over all space, we
obtain
d P
g
+ SP
=
dt
T0

(6.30)

where the angle brackets denote a spatial integral, and SP represents a surface integral
term. Ignoring the surface term for the moment, it is evident that P is very closely related
to the vertical momentum integral of the previous section. In fact, far from a solid
boundary, P satisfies exactly the same evolution equation as w and can therefore be
identified with the integrated vertical momentum.
In order to use (6.30) as a basis for the dense gas puff dynamics, we must
represent the surface integral term and also relate the puff motions to the vorticity
moment integral. Since the velocity field strictly vanishes at a solid boundary, the
surface integral SP involves only the viscous diffusion terms; we therefore represent it as
a drag term, whose detailed specification will be described below. The relationship
between the vorticity moment integral and the puff dynamics is based on a simple shape
assumption for the induced velocity field, since the relation between velocity and
vorticity is linear. This allows us to superpose the velocity fields from neighboring puffs
by simple addition, and represent the interactions between a collection of dense puffs.
The vorticity moment evolution equation (6.30) also depends linearly on the buoyancy
integral, so that the total vorticity moment can be represented by summing the moments
of individual puffs. The puff evolution is still nonlinear since the dynamic fields are
composed of the summation over all puffs, but the puff moments evolve independently
and the basis of the dynamics is contained in the shape assumption for the velocity field.
In general, the velocity field induced by a dense gas cloud involves an outward
radial flux in the cloud itself, spreading the material horizontally over the ground surface.
This horizontal divergence is accompanied by a vertical velocity gradient as material
moves downward toward the ground. A schematic cross-section is illustrated in Figure
61. This flow field is associated with azimuthal vorticity, and we can obtain a general
relation between the velocity magnitude and the vorticity moment.

84

DRAFT

Figure 6-1. Schematic illustration of slumping dense cloud.

We further simplify the representation by assuming that the vorticity is

concentrated in a horizontal sheet where the velocity is discontinuous, as illustrated in
Figure 6-2. The horizontal velocity components, u and v, are independent of z below the
sheet, and are zero above. The horizontal velocity field below the discontinuity is then
used to move the scalar field. The vertical velocity is proportional to z, and is determined
by the horizontal divergence. This ignores the return circulation outside the dense gas
cloud, but captures the essential features of the motion of the dispersing contaminant.

u = 0 , w = w x, y

a f

u = u x, y , w = z H u

a f

Figure 6-2. Simplified velocity field representation for a dense cloud.

If we assume that the scale of the horizontal radial velocity is U0, the radial
dimension is LH, and the vertical dimension is LV, then the vorticity magnitude will be
U 0 LV . This assumption neglects the horizontal gradient of the vertical velocity
component in the definition of vorticity, but this contribution is negligible when the cloud
is shallow in comparison with its width. An order of magnitude estimate for the vorticity

85

DRAFT
moment is thus obtained as U 0 L3 , using LH as the magnitude of the radial vector, x, and
H
2
LH LV as the volume estimate.
A simple assumption for the radial velocity field uses the Gaussian form to give

FG

r
r2
exp 2
LH
LH

ur = U 0

IJ

(6.31)

This satisfies the linear gradient requirement at the center of the puff and vanishes at large
distances. We cannot assume circular symmetry, however, since SCIPUFF uses a
generalized Gaussian description and the velocity shape assumption must be applied to
elliptical puffs. When a puff is non-circular, the outflow along the longer axis of the
ellipse is smaller than that along the shorter axis, since the pressure gradient driving the
flow is smaller. If we weight the outflow using the inverse of the axis length, then the
radial distribution given in (6.31) can be generalized to an elliptical shape as

FG

IJ

X LY
X2 Y2
u X = U0
exp 2 2
L X LH
L X LY
Y LX
LY LH

uY = U 0

H
F X
expG
H L

2
X

Y2
L2
Y

(6.32)

K
IJ

(6.33)

where X and Y are the principal axis coordinates in the horizontal plane with the puff
centroid as origin, and 2 L2 = L2 + L2 . The length scales LX and LY are related to the
H
X
Y
semi-major axes of the puff as

L X ,Y = Cd1

FG

xx

+ yy
2

1
2

xx

yy

+ 4 2
xy

IJ

(6.34)

and the empirical coefficient Cd1 is chosen to be 1.5. Note that the velocity components in
(6.32)-(6.33) must be rotated from the frame of the principal axes into the (x, y)
calculation coordinates. Horizontal velocity gradients are computed by differentiating and
rotating the four components appropriately. In the principal axis coordinates, the velocity
gradients are
u X
X 2 U 0 LY
X2 Y2
exp 2 2
= 1 2 2
X
L X L X LH
L X LY

FG
H

IJ
K

uY
Y 2 U0 LX
= 1 2 2
Y
LY LY LH

FG

IJ

FG

H
F X
expG
H L

2
X

86

Y2
L2
Y

IJ
K

IJ

(6.35)

(6.36)

DRAFT

U
u X
X2 Y2
= 2 X Y 0 exp 2 2
Y
LH
LX LY

(6.37)

U
uY
X2 Y2
= 2 X Y 0 exp 2 2
LH
X
LX LY

(6.38)

The scaling velocity, U0, is related to the vorticity moment, which we identify
with w . Integrating the moment of the vorticity based on vertical gradients of the
velocity field (6.32)-(6.33), we obtain
w

U0 =

LH
L L2
Y

(6.39)

2
X

Specification of the velocity field of an individual puff leads to the evolution

equations for the puff moments, using both the velocity and the velocity gradients. These
quantities are calculated at the puff centroid location by summing the contributions from
all the overlapping puffs. Note that a single puff has no mean velocity at its centroid,
since the components (6.32)-(6.33) vanish at the centroid, where X = Y = 0. Thus, the
total dynamic velocity perturbation from dense gas interactions is

u x x

ud =

b g

(6.40)

where the sum is taken over all dense gas puffs, and the summand represents the velocity
field (6.32)-(6.33), after rotation into the project coordinate frame, evaluated at the
centroid of puff- . The other velocity component and all the horizontal gradients are

computed in a similar manner.

The dynamically-induced velocity and velocity gradients are simply added to the
ambient flow field to produce the total transport and distortion of the puffs. The vertical
velocity gradient is defined from the sum of the two horizontal divergence components,
maintaining a total divergence of zero. The vertical component of velocity is defined
from the velocity gradient, assuming zero as the surface boundary condition. Thus


wd
ud vd
=
+
z
x
y


(6.41)

where the subscript d denotes the overlap sum from all contributing puffs with dense
dynamics. The vertical velocity is defined as

87

DRAFT

wd = z

wd
z
exp
z
cV

(6.42)

where z is the puff centroid height, and cV is a representative vertical scale height for
the puff. The exponential decay factor is introduced to reduce the velocity for puffs
further from the surface.
The dense gas effects are only calculated for puffs that interact with the ground
surface. Negatively buoyant puffs remote from the surface will simply fall downward
under the buoyancy forcing described in Section 6.1.1. Puffs are judged to interact with
the ground when they have negative vertical momentum, w p c , and their centroid is less
than 2 cV above the ground. The first requirement ensures that they are descending, and

the second checks that they are in contact with the ground. When these conditions are
satisfied, the ambient temperature gradient term in the buoyancy equations, (6.2) and
(6.5), is neglected, since the vertical velocity integral is interpreted as a vorticity moment
and there is no vertical transport at the solid surface.

6.2.2 Turbulent Entrainment.

The presence of the ground introduces stability effects on the turbulence in

addition to the pressure gradient effects of the mean velocity field. In a free atmosphere,
any density perturbation forms a stable vertical gradient on one side and unstable on the
other. A free dense cloud is unstable on the lower side, while a positively buoyant
perturbation is unstable on the upper side. However, when the density perturbation is
constrained by the ground surface, a stable gradient is established throughout the cloud.
The density gradient suppresses vertical turbulent diffusion since vertical motions
transfer kinetic energy to potential energy as the density perturbations are mixed. This
mechanism applies to the ambient turbulent motions as well as the dynamically-induced
motions in the slumping dense cloud.
When a puff is in dense ground effect, as determined by the ground interaction
criterion in the previous section, a dense cloud turbulence entrainment model is applied.
The framework is similar to the buoyant rise parameterization in Section 6.1.3, but we
use different velocity, length scales, and stability dependence.

88

DRAFT
The internal velocity scale for a dense gas puff is based on the lateral velocities
since the vertical motion is constrained. Similarly, we base the length scale on the
vertical scale of the cloud since the turbulent eddies are also constrained by the solid
boundary. The internal turbulent velocity scale is given by
2
q d = c q1 Fd2 Rid Vd2

b g

(6.43)

where
V = u + v + ( xx + yy )
2
d

Fd =

2
d

2
d

( )
wd
z

(6.44)
(6.45)

1 + Cd 2 Rid

and
Rid =

g d p T0 bp

(6.46)

Vd2 + V 2

The dynamic velocity scale Vd represents the lateral 'slumping' motion in the
dense cloud. The gradient term is required to provide the velocity scale for the selfinduced motions in a puff, since the mean self-induced velocity is zero at the puff
centroid. The internal dynamic scale, d, is defined as the vertical puff scale, cV, since
the vertical depth of the dense cloud characterizes the shear-generation scale. We note
that the Richardson number is based on the total velocity, because the balance between
buoyancy and shear in the cloud determines the effective stability and ambient motions
are included in the balance. The coefficient in (6.45), Cd2, is chosen to be 25 on the basis
of comparisons with laboratory data.

In addition to the stability effect on the internally generated turbulence, the

formation of a stable layer at the ground affects the ambient turbulence and also the
ambient velocity profile. Since boundary layer turbulence is principally driven by
interaction with the ground surface, the reduction of vertical transport through the dense
cloud layer can significantly modify the local profiles. We represent this effect by
applying the damping factor Fd to the equilibrium eddy diffusivity, K zeq , defined in
(2.31).

89

DRAFT
The mean ambient velocity is also reduced, due to the suppression of vertical
mixing, using a factor
0.7 + 0.3Fd
as suggested by Puttock (1988), based on field experiments. This simple factor gives a
maximum speed reduction of 30% when the Richardson number is very high.
The turbulent diffusivity, dissipation timescale and internal horizontal scale
equation are modified as in (6.25)-(6.27), but using qd and d in place of qp and p.
Additionally, the vertical compression effect is included in the equation for the internal
vertical scale, cV.

6.2.3 Surface Drag.

As the dense cloud moves over the ground surface, the frictional drag of the rough
surface tends to reduce the lateral slumping rate. This effect is represented by the surface
integral terms in (6.30). We represent the drag effect through a damping coefficient on
the vertical momentum (or vorticity moment) integrals, w p (or P ) and w p c . In
addition, the interaction with the ground introduces a heat flux directly into the cloud, and
we assume that this exchange rate occurs at the same rate as the momentum exchange.
The damping is therefore applied also to the temperature perturbation integrals.
The surface time scale is based on the roughness length and a wall-layer
assumption. If we define an inverse time scale as

w1 =

1 u b
u b t

(6.47)

and assume that

ub
=
t
z

where

(6.48)

is the turbulent momentum flux, then using the height scale of the cloud and the

wall-layer relation between the surface stress and local wind speed, we obtain

d
z0

FH

w1 = log 1 +

IK

V 2 + Vd2
d

90

(6.49)

DRAFT
6.2.4 Terrain Effects.

In the presence of sloping terrain, a dense cloud will tend to move down the slope
under the influence of the gravitational forcing. A simplified description of this effect
can be included in the model using an equilibrium relation between the buoyancy force
and the surface drag to provide a horizontal velocity perturbation. Since the vertical
momentum is simply an integral of the balance between the buoyancy force and the
surface drag, as in (6.30), we further assume that the terrain-induced horizontal motion is
determined from the relations

ud = w p
vd = w p

hx

(6.50)

2
1 + hx

hy

(6.51)

2
1 + hy

where w p is defined by (6.9).

The terrain-induced velocity is added to the dynamic velocity, (ud, vd) computed
from the puff interactions as in (6.40). The additional terrain component is used to move
the centroid, and also appears in the turbulent entrainment and surface drag terms.

6.3 FLASHING LIQUID/AEROSOL EFFECTS.

6.3.1 Volatile Liquid Specification.

Volatile materials such as ammonia and chlorine are often stored at ambient
temperature in pressurized vessels; typically the overpressure in such vessels may be as
much as 10bar. A sudden rupture can result in a nearly instantaneous release while a
puncture in the vessel can result in a continuous jet of contaminant. In either case, the
release will be violent and take the form of a flashing liquid droplet-vapor mixture as the
material boils explosively and the pressure reduces to ambient as the mixture cools to its
boiling point. Further cooling can occur as the volatile liquid vaporizes. As a result the
ensuing cloud will typically have a density (over and above any molecular weight effects)
much higher than the ambient air due to the presence of liquid aerosol and the low
temperature. Such density differences greatly affect the dispersion of the contaminant, as

91

DRAFT
discussed in the previous section. In this section, we describe the implementation of a
two-phase aerosol cloud model in SCIPUFF so that the dispersion of a flashing, volatile
liquid release can be represented properly.
A homogeneous, thermodynamic equilibrium approximation is used to determine
the state of the contaminant. This is appropriate for small liquid droplets, which respond
very quickly to changes in their environment, and greatly simplifies the model. The
model tracks five components within a contaminant cloud; contaminant liquid and vapor,
water liquid and vapor, and dry air. The latent heat of atmospheric water can contribute
significantly to the thermodynamics of the cloud, and it is therefore important to include
water in the model.
Within SCIPUFF, the flashing aerosol saturation vapor pressure is specified in
the same way as that of a liquid material, i.e., using the Antoine coefficients as described
in Section 5.1. For liquid materials, we assume that the volatility is low so that the
droplet evaporation process in Section 5.2.1 ignores the vapor pressure in the surrounding
atmosphere. For the flashing aerosol, we have the opposite situation, where the local
vapor pressure completely controls the rate of evaporation.

6.3.2 Homogeneous Equilibrium Model.

The homogeneous equilibrium (HE) model assumes the contaminant cloud

aerosol is distributed in a self-similar manner to the other cloud variables (concentration,
temperature perturbation, etc.). The liquid aerosol and vapor are taken to be in
thermodynamic equilibrium, allowing the cloud to be characterized by a single
temperature. The cloud description involves five components, contaminant liquid and
vapor, water liquid and vapor and dry air and is analogous to the model used in SLAB
(Ermak, 1990). A typical release scenario would be the sudden venting of a volatile
liquid from a pressured vessel. The rapid release and flashing are treated as an adiabatic
process with the contaminant in liquid and vapor phases at a temperature equal to the
normal boiling point. The mass fraction of material in the liquid phase depends on the
superheat (ambient temperature excess above normal atmospheric pressure boiling point)
and the latent heat of vaporization. We find that materials such as ammonia, hydrogen
fluoride and butane have liquid mass fractions of 60 to 80% when released into standard
temperature atmospheres. The resulting cloud is thereby mostly liquid by mass, and

92

DRAFT
vapor by volume. As ambient air is entrained into the cloud, the partial pressure of
contaminant decreases, the aerosol vaporizes and cools below normal boiling point. If
ambient water vapor is present, it can condense, release latent heat and reduce the
temperature drop due to contaminant vaporization. Water is only considered as liquid or
vapor; ice formation is neglected. Ultimately, enough ambient air will be entrained so
that all liquid contaminant and water are vaporized.
The HE model will be a good approximation for small liquid droplets. Typical
thermal response times for liquid droplets are of order 100s milliseconds for diameters
of order 100s microns. If the release energy is sufficient enough to create droplets in the
100 micron or less size range, the vapor temperature will be approximately equal to that
of the droplet. Also, the droplet fall velocity will be small, of order 10s cm/sec, and thus
droplets are unlikely to rain out of the cloud. Kukkonen et al. (1994) compared the HE
model with a non-equilibrium droplet model for an instantaneous release of ammonia,
and found nearly identical results for both models for droplets less than 100 microns.
Non-ideal solution effects were also considered and found to have minor effects on cloud
evolution.
The physical picture for the HE model can understood by considering a uniformly
mixed, homogeneous cloud of mass, M, volume V and density m at time t. The
composition of the cloud is given by mass fractions, mlc for liquid contaminant, mvc for
vapor contaminant, mlw for liquid water, mvw for vapor water, mda for dry air, together
with the cloud temperature, T. Pure component densities are given as lc, etc., and the
overall cloud density is m, which is defined as

m1 = mvc vc1 + mlc lc1 + ma a 1

(6.52)

where a is the ambient air density. In this expression, we are assuming that the
condensation of water does not influence the cloud density, since the water content of the
air component is not changed. Also, atmospheric moisture content is generally small
enough that the standard variation of air density with altitude and temperature can be used.
Thus, cloud density is only influenced by the contaminant and its phase partition.
During a time interval, t, a volume V of mass M of ambient air is mixed into
the cloud. This results in a new set of mass fractions, density and temperature, denoted
by primes.

93

DRAFT
Mass conservation gives the following relations:

m (1 + r ) = m + a r

Total mass:

(6.53)

Contaminant mass:

m (mlc + mvc ) (1 + r ) = m (mlc + mvc )

(6.54)

Water mass:

m (mlw + mvw ) (1 + r ) = m (mlw + mvw ) + m a r

(6.55)

m mda (1 + r ) = m mda + (1 m) a r

Dry air mass:

(6.56)

where a is the ambient air density, m is mass fraction of ambient water vapor and r is the
entrainment mixing ratio, given by
r=

V
V

i.e., V = (1 + r ) V
The sum of the mass fractions is unity by definition, i.e.,
mlc + mvc + mlw + mvw + mda = 1

(6.57)

and similarly

mlc + mvc + mlw + mvw + mda = 1

so that the total mass equation, (6.53), can be obtained from summation of the three
component equations for contaminant, water, and dry air, (6.54)-(6.56).
Conservation of energy is given by the following.

m V cm T = m V cm T + r a V ca Ta + ( m V mlc m V mlc ) Lc

+ ( m V mlw m V mlw ) Lw

(6.58)

where cm is the specific heat of the mixture, ca is the specific heat of the ambient air, and
Lc and Lw are latent heats of vaporization for contaminant and water, respectively. The
cloud-average specific heat is defined as the mass-weighted average of the component
specific heats, i.e.,
cm = mvc cvc + mlc clc + mvw cvw + mlw clw + mda cda

(6.59)

When liquid and vapor components are present, the component vapor pressure
just above the droplet surface is the same as at the droplet surface. The homogeneous
equilibrium assumption implies that the vapor pressure of each component is the same
everywhere in the cloud, and must therefore be equal to the saturation pressure. For

94

DRAFT
separate, pure component droplets, the mass vapor fraction of the contaminant
component is thus
mvc =

Pcs ( T ) M c

(6.60)

Pa M v

where Pcs is the saturation vapor pressure at the cloud temperature, T, and Pa is the
ambient pressure. Mc is the molecular weight of the contaminant, and Mv is the molecular
weight of the cloud vapor mixture, given by

M v1 = mvc M vc1 + mvw M vw + mda M da1

(6.61)

The above set of relations can be reduced to a set of three non-linear coupled

equations for mlc , mlw and T . For given initial mass fractions and temperature, new
values can be computed for a given entrainment ratio. r. At each time step in the puff
(cloud) evolution, r is estimated by the reduction in total contaminant concentration, and
used to compute the new thermodynamic equilibrium state. This non-linear system of
equations is solved using the Newton-Raphson technique.

6.3.3 SCIPUFF Implementation.

The homogeneous equilibrium equations are written in terms of local point values
of the thermodynamic variables. The puff model uses average values for each puff to
determine a single description in terms of liquid fraction and temperature. At each
timestep the puff temperature is determined from the nonlinear correlation, p c , where
p is the temperature perturbation from the ambient value. The initial temperature for
the thermodynamic calculations in 6.3.2 is then estimated as
T = Ta +

pc

(6.62)

where Ta is the local ambient temperature.

Note that the total correlation is used in the definition of the temperature, so that
the turbulent fluctuations are also included. This is effectively defining a conditionally
averaged temperature associated with the concentration field, c, which is the most
appropriate measure for the equilibrium calculation.

95

DRAFT
In addition to the temperature correlation, a flashing aerosol puff stores the
contaminant liquid mass fraction, and also the water mass fraction. These three values
define the initial thermodynamic state, from which the equilibrium state is solved as
described above. The calculation also requires the entrainment factor, r, which is defined
from the reduction in contaminant concentration from the previous step. As for
temperature, the conditional concentration is used, so that
r=

c2
old

(6.63)

new

where old and new refer to the previous and current timestep values, respectively.
Following the equilibrium adjustment, new values are obtained for the
temperature, T, and liquid mass fractions. The new mass fractions are simply stored for
the next timestep, but the integrated puff temperature variables must be adjusted to reflect
the thermodynamic energy input associated with the evaporation process. The new
temperature correlation is defined as

pc

final

= c (T Ta )

(6.64)

where T is the equilibrium temperature solution. The total temperature perturbation

associated with the puff is modified to give

final

cm p + H c

cm

(6.65)

where the specific heat release per unit mass of contaminant is defined as

H = m

mcl mcl
m
m

Lc + wl wl Lw
mc mc
mc
mc

(6.66)

Here mc = mcl + mcv is the total contaminant mass fraction, and m is the updated average

cloud density from (6.52). The mean temperature correlation is similarly adjusted to give

p c

final

cm p c + H c 2

cm

(6.67)

The flashing aerosol also introduces dynamic effects through the temperature
and phase changes. The treatment of puff buoyancy is described in Sections 6.1 and 6.2,

96

DRAFT
and the flashing aerosol puffs use this same methodology but now the contaminant
density includes the liquid aerosol component in addition to the vapor density. Thus, the
buoyancy factor in (6.11) also uses the mean density, m, from (6.52).

97

DRAFT
SECTION 7
SOURCE SPECIFICATION
7.1 CONTINUOUS SOURCES.
A continuous source is specified as a constant mass release rate, RQ , for a finite
duration period, TR, starting at time, tR. The material type and the release location,
x R = x R , y R , z R , must also be defined. The source geometry is simply defined in terms
of the spread parameter, R.

At each model time step during the active period of the release, t R t < t R + TR , a
new puff is initialized and added to the existing list. If the model time step is t, then the
mass of the new puff is readily determined as
Q = RQ t

(7.1)

The centroid location of the new puff accounts for the passive transport by the
mean wind, giving
1
x = x R + u t
2

(7.2)

which represents the center of mass of a plume segment of duration t originating at the
source position.
The second-moment tensor also accounts for the mean transport, so that

1
4

ij = 2 ij + ui u j t 2
R

(7.3)

The addition of the mean velocity terms in (7.3) produces an elongation of the puff in the
downstream direction, and a smooth overlap between consecutive puffs.
The turbulent flux correlation moments, x i u c , are all initialized to zero, since
j
there is no flux at the source, and the mean-square concentration integral is

98

DRAFT
c

2
RQ t

4 2 V
R

(7.4)

V 2 = ui ui + u 2 + v 2

(7.5)

where

The initial concentration fluctuation variance is thus effectively set to zero.

The internal fluctuation length scales, c and cV, are set equal to R, but the
second horizontal scale, cH, reflects the continuous 'plume' nature of the release. Thus

FH

cH = min y ,

tR

V dt

IK

(7.6)

where y is the composite horizontal turbulence scale defined in (2.92) and the integral
represents the length of the plume.

7.2 INSTANTANEOUS SOURCES.

Instantaneous sources only involve a single puff creation stage, and are specified
by a release time, tR, a release location xR, and a CLOUDTRANS puff file (Henn et al.,
1995). When the SCIPUFF calculation reaches the release time, the puffs in the
CLOUDTRANS file are released into the dispersion domain. A collection of puffs can
be released in a single source event to describe an arbitrary spatial distribution, and this
input is translated directly into initial conditions for each puff. The CLOUDTRANS
input must contain material identifiers that correspond to existing SCIPUFF material
types, and must also provide mass, centroid location, and spread parameters for each
puff.
The centroid location for each puff is used as a three-dimensional offset from the
general release location, xR, and the puff moments ij, can be specified directly. The
CLOUDTRANS file can also neglect the off-diagonal moments by specifying x, y and
z, or simply . In the latter case, a spherical puff will be released.
As for the continuous release, the turbulent flux moments are initialized to zero.
The mean-square concentration integral is calculated using the interaction integrals in
Section 2.2 for all the puffs in the release. This strictly gives the square of the mean, but

99

DRAFT
this is equal to the mean square under the assumption of zero fluctuation variance. The
internal fluctuation scales are initialized using the puff spread moments if the scales are
not directly provided on the CLOUDTRANS file. Currently, SCIPUFF uses
c = min x , y
cH

c
= maxc

, y

(7.7)

cV = z

7.3 LIQUID POOL EVAPORATION.

7.3.1 Pool Evaporation.

SCIPUFF models evaporation from a liquid pool using mass transfer theory with
consideration for the different modes of heat transfer. The model is very similar to the
evaporation models described in Section 5, and the same notation will be used here. The
Sherwood number is the dimensionless measure of the mass transfer rate, and the
appropriate form in this case is
m
DW Cs C

Sh =

(7.8)

where D is the diffusivity of the vapor in the surrounding air, W is the width of an
equivalent square pool, Cs is the droplet saturation concentration, and C is the ambient
vapor concentration. We neglect the ambient concentration in (7.8), so that the pool
source model is valid only for low volatility, where the air concentrations are reduced
rapidly by advection. The saturation concentration is taken at the pool temperature. We
use expressions from the HGSYSTEM model suite (MacFarlane et al., 1990) to
approximate an equation for the Sherwood number:

R0.664Sc
Sh = S
T 0.037Sc

Re 1 2
, Re W < 3.2 10 5
W
13
4
(Re W5 15200) , Re W 3.2 10 5
13

(7.9)

and

Re W = a UW a

(7.10)

where U is the 10 meter wind speed, a is the air density and a is the air viscosity. The
length scale is defined as

100

DRAFT
W = 1 2 Rp

(7.11)

where Rp is the pool radius.

The droplet saturation concentration is a function of the pool temperature, which
is obtained from an energy balance method. The latent heat loss associated with
evaporation is balanced by the heat transfer from convection, q c , and conduction to the
ground, q G ; we assume that radiation losses are negligible. If is the latent heat of
evaporation, we obtain

m = qc + qG

(7.12)

By analogy with the mass transfer in (7.8), the convective heat transfer is modeled
using the Nusselt number, Nu , as
q c = Nu a W (T Tp )

(7.13)

where a is the thermal conductivity of air, T is the 10 meter air temperature, and Tp is
the pool temperature. The Nusselt number is defined similarly to the Sherwood number in
(7.9) but uses the Prandtl number for air in place of the Schmidt number.
The ground conduction estimate is based on the semi-infinite slab model (Raj and
Morris, 1987)

G W 2 (TG Tp )
qG =
G t p

(7.14)

where TG is the ground temperature, G is the ground thermal conductivity, G is the

thermal diffusivity, and t p is the age of the pool. The physical properties of the
underlying ground surface are taken to be representative of sand; the thermal conductivity
is assumed to be 0.27 Wm1K1 and the diffusivity is 2.2107 m2s1 (Incropera and
DeWitt, 1990)
Substituting the two heat flux estimates into (7.12), the pool temperature is
obtained as
Nu a T + CG G TG m W
Nu a + CG G

Tp =

where

101

(7.15)

DRAFT
CG =

(7.16)

G t

7.3.2 Vapor Puff Formation.

Vapor puffs form at the surface as the liquid pool evaporates through diffusion
and convection. Once these puffs develop, they translate and diffuse like other gas puffs.
Like their primary vapor puff counterparts, their centroid location, size, mass, and
concentration variance must be initialized.
Liquid pools are treated as continuous sources, but with a time-dependent mass
release rate defined by the evaporation model described in the previous section. The
horizontal spread parameters are initially set to the source radius, and the vertical spread
is determined from the local turbulent diffusion rate. A simplified estimate based on the
surface friction velocity is employed, so that

z = u* t

(7.17)

where is von Krmns constant.

The pool source mass is depleted at each time step, and the release is terminated
when the mass is exhausted. The pool area is assumed to remain constant throughout the
release.

7.4 FLASHING LIQUID/AEROSOL SOURCES.

Only instantaneous sources of flashing aerosol materials are supported. The
release is calculated from the mass of flashing contaminant, together with a specification
of the entrainment fraction. We assume that the liquid contaminant is stored at the
ambient temperature under pressure, which is instantaneously reduced to the local
atmospheric value. First, the thermodynamic state of the pure material is determined
using the assumption that the cloud will rapidly reach the contaminant boiling point as
the liquid is evaporated. Thus, assuming that the heat of vaporization balances the
cooling of the liquid fraction, the estimate of the initial vapor mass fraction is given by

102

DRAFT
clc
(Ta Tboil )
Lc

mvc =

(7.18)

where Lc is the latent heat of evaporation of the contaminant, and clc is the liquid specific
heat (see Section 6.3). The evaporation determines the effective initial size of the release,
prior to the entrainment of the ambient air. Using the assumption of pure material to
determine the density as

m1 = mvc vc1 + mlc lc1

(7.19)

the initial radius of the spherical cloud is simply

1

(7.20)

3
4

r0 = Q0

1
m

where Q0 is release mass. The initial Gaussian spread is set to one half of this radius.
Next, we assume an entrainment of the ambient air, based on entrainment ratios
for each of the three coordinate directions, i.e., the three spherical Gaussian spread
parameters are multiplied by an entrainment factor. Since the entrainment is a complex
function of the release mechanism, the entrainment factors must be specified as inputs
(see Section 13.1.3). Thus, xx = Ex2 r02 where Ex is the x-direction entrainment factor,
and similarly for the y- and z-directions.
Having mixed the contaminant cloud with ambient air through the entrainment
process, a new thermodynamic equilibrium state is determined using the model described
in Section 6.3.

103

DRAFT
SECTION 8
SPECIAL MATERIALS
8.1 ACTIVITY DECAY.
Activity decay can be applied to any material in SCIPUFF to represent a simple
reduction in viable fraction (for a biological agent) or other linear decay effects. Each
puff carries an activity coefficient, fA, which is initialized to unity at release time. During
the transport calculation, the coefficient evolves according to the relation
df A
= k A f A
dt

(8.1)

where kA is the user-specified decay rate. The decay rate specification includes a diurnal
variation, so that effects that depend on solar radiation can be described. The actual rate is
obtained from the daytime maximum, k Amax , and nocturnal minimum, k Amin , as

af

k A = max k Amax sin , k Amin

(8.2)

where is the solar elevation angle.

All deposition and dosage integrals use the activity-weighted concentration, fA c,
so that deposition, for example, is computed as

D x, y, t =

f A v D c x, y,0, t dt

(8.3)

The decay calculation continues after material has been deposited on the surface,
so that

D
= k A D
t

(8.4)

This assumes that the decay process is equally effective for material on the surface and in
the air. The variance of the deposition decays at twice the rate, i.e.,

2
D = 2 k A 2
d
t

(8.5)

104

DRAFT
8.2 NUCLEAR WEAPON MATERIALS.
Nuclear weapon materials are a special category of material used in conjunction
with the NWPN source module. The source module provides an initial particulate cloud
together with normalized radioactivity as the mass variable for the puffs. The activity is
defined at 1 hour after release, and is carried as a conserved variable in the dispersion
calculation. Radioactive dose and dose rates are estimated from the surface deposition of
the material using empirical decay laws from the NEWFALL code (McGahan, 1990).
The actual radiation activity is determined from the 1 hour activity using the decay factor

f R = t tR

(8.6)

where t is the current time and tR is the release time in hours, and p is the decay exponent
defined as a material property. This factor multiples the conserved 1 hour activity, c,
carried by the puffs
Since SCIPUFF can describe multiple sources, we must consider the possibility
that several weapons may be released at different times. However, the deposition field is
calculated as a continuous accumulation from any depositing puffs, so the deposited
activity cannot be identified with a particular weapon source. We therefore associate a
mass-weighted release time with the activity in any grid cell, so that an average time is
used for the calculation in (8.6). This requires each puff to carry a release time, so that
the deposition integration can also calculate
t

D x, y, t = t R v D c x , y,0, t dt

(8.7)

The average release time is then defined as

tR =

D
D

(8.8)

In addition to the mass deposition, an integrated radiation dose is calculated for

NWPN materials. The dose rate from the deposition field is defined using the decay
factor (8.6) as
R = fR D

(8.9)

The mean radiation dose is therefore calculated as

105

DRAFT
t

DR =

f R t D t dt

af af

(8.10)

The variance of the dose is obtained from the relation

t t

DR =

zz

f R t f R t t t dt dt
D
D

af a f af a f

(8.11)

0 0

and using the definition of the deposition in (2.99) we obtain

t

DR2 = dt dt d 1 d 2 f R t f R t v D 1 v D 2 c 1 c 2

z z z z
0

af a f b g b g b g b g

(8.12)

Using our usual definition of the autocorrelation function for the concentration
fluctuations, this expression is written
t

DR2 = dt dt d 1 d 2 f R t f R t v D 1 v D 2 c 2 ( 1 ) Rc 2 1

z z z z
0

af a f b g b g

(8.13)

where Rc is the correlation function. Neglecting the variation in deposition velocity, at

least over the correlation timescale for the concentration fluctuations, gives
t

2
DR2 = dt dt d 1 f R t f R t v D 1 c 2 ( 1 )

z z z
0

af a f b g

t 1

d Rc

af

(8.14)

A simplified model for the integral of the correlation function is given by

Rc ( ) d = 2Tc H u

(8.15)

where H is the Heaviside step function, and Tc is the integral timescale. This simply
switches the correlation integral to the fully infinite integral value when the upper bound
exceeds zero. In reality the integral will increase from zero to the maximum value over a
range of order Tc, and more complex expressions could be considered. The step function
simplifies the integrals dramatically, however, giving

106

DRAFT
t

2
DR = dt dt d 1 f R t f R t v D 1 c 2 ( 1 ) 2Tc 1 H t 1

z z z
z z z
z z
0

b g b

af a f b g

2
= dt dt d f R t f R t v D c 2 ( ) 2Tc
t

af a f af
t

a f a f z d v a f

= 2 dt dt f R t f R t
0

af

2
D

c 2 ( ) 2Tc

af

Now the innermost integral is simply the variance in the mass deposition, so we finally
obtain
t t

DR2 = 2

zz

f R t f R t 2 t dt dt
D

af a f

a f

(8.16)

0 0

and the computation requires that we store an integral of the mass deposition variance,
t

VR t =

af

f R t 2 t dt
D

af

af

(8.17)

which is then integrated again to give the dose variance as

t

DR = 2 VR t dt

af

(8.18)

8.3 NUCLEAR FACILITY MATERIALS.

8.3.1 Radiation Dose Factors.

Nuclear facility materials are a second special category of material used in

conjunction with the NFAC source module. The nuclear reactor radiation source and
dose calculation is based on the Nuclear Regulatory Commission's RASCAL code (Athey
et al., 1993) extended to a worldwide inventory. The NFAC source module provides
radiative dose factors to convert SCIPUFF dispersion predictions for a unit mass release
into the various human effects estimates. The capability has been generalized in this
version of SCIPUFF, so that multiple materials can be defined including particle and
liquid types and allowing time-dependent releases to be described.
The generalized approach defines a set of Component radiation doses that are
stored on the surface files. Additionally, a set of Plotted doses is defined, each one

107

DRAFT
being a weighted sum of the Component doses, and these are the radiation doses and
dose rates available as output display quantities. Suppose the component dose rates are
Ri where i ranges from 1 to NC, the total number of component doses. For each NFAC
material, a dose factor is defined for each component dose and this factor is specified as a
function of time on a separate input file. This input file, with file extension RAD, is
described in Section 13.1.5 below.
Component doses comprise three types, namely air doses, ground doses, and the
cloudshine dose. Air doses are based on the local surface concentration of material, and
are appropriate for inhalation effects, while ground doses are based on the deposition of
material. Cloudshine is a special dose calculated from the remote radiation from the
three-dimensional concentration field, and is described separately in the next section.
Thus, for an air dose, the local component dose rate at position ( x, y ) and time t is given
as

Ri ( x, y, t ) =

ij ( t ) c j ( x, y, zs , t )

(8.19)

where the sum is taken over all NFAC materials and ij is the dose rate factor for
contribution from NFAC material-j to component dose-i at time t. Currently,
contributions from different NFAC materials are treated as independent, since
dispersion model does not calculate correlations between different materials. Thus,
dose rate variance is obtained from the summation of each material contribution,

ij2 c j 2

Ri2 =

the
the
the
the

(8.20)

while the mean dose rate is directly obtained from the mean concentrations. The
integrated doses are stored on the adaptive surface grid, using the same methodology as
the dosage calculations described in Sections 2.4.1 and 3.7. Thus, the mean dose is
defined as
t

Di =

ij ( t ) c j ( t ) dt

(8.21)

ij2 ( t ) c j 2 ( t ) Tc( j ) ( t ) dt

(8.22)

and the dose variance is

t

Di =
2

where Tc( j ) is the autocorrelation timescale for NFAC material-j.

108

DRAFT
Ground dose components base the dose rate on the material deposition fields, so
that

Ri ( x, y, t ) =

(
ij ( t ) Dj ) ( x, y, t )

(8.23)

(
where Dj ) is the deposition field for NFAC material-j. Integrated ground doses are stored

on the adaptive deposition grid from the integration of the dose rate in (8.23), thus the
mean ground dose component is given by
t

Di =

(
ij ( t ) Dj ) ( t ) dt

(8.24)

The ground dose variance is calculated using the assumption that the
contributions from each NFAC material on the same adaptive grid cell are perfectly
correlated. However, summation rules for the adaptive grid assume independence of the
variance when contributions are added from different grid scales. Thus, the integral of
the ground dose variance is given by

ij ( t ) D
j





Vi =

( j)2

dt

(8.25)

and the dose variance is then

t

Di2 = 2 Vi ( t ) dt

(8.26)

as in (8.18).
Following the dispersion calculation, the output display consists of Plotted
doses, defined as weighted sums of the Component doses. Thus, a Plotted dose, Pi, is
defined as

Pi =

ij D j

(8.27)

where the summation is over the Component doses, and the s are constants. The
variance of Pi is obtained from the square of the weighted sum of the standard
deviations in the Component doses. This effectively assumes that the fluctuations in each
of the components are perfectly correlated.

109

DRAFT
The generalized methodology provides an improved treatment of the timedependent dose factors, since the actual dose rate is integrated on the surface files. This
compares with the previous methodology where an average dose factor was used to
estimate doses from integrals of the material concentration and deposition fields. In
addition, the new scheme allows the dose rates to be displayed since they can be
constructed directly from the local fields using the appropriate dose rate factors.

8.3.2 Cloudshine Dose.

The radiative dose factors described in the previous section are applicable to
direct inhalation of radioactive nuclides or for close-range exposure to the material either
deposited on the surface or locally present in the atmosphere. The gamma radiation from
airborne nuclear materials is more difficult to estimate, since doses can be accumulated
from remote material and cannot be determined solely from the surface integral fields, in
general.
The gamma cloudshine dose rate at any point on the surface is strictly defined as a
three-dimensional integral over the entire spatial volume distribution of radiating
material. The dose rate must be calculated at each timestep and at each point on the
surface, which can increase the computation time significantly. Two alternative methods
are therefore provided for estimating the gamma cloudshine dose. A "submersion"
assumption can be specified, in which the surface concentration is used in a semi-infinite
cloud approximation, or a more complete treatment of the remote shine dose integrals can
be used. The submersion approximation requires no additional computation, since the
dose is calculated from the time-integrated surface concentration using the pre-calculated
dose factors, exactly as described in the previous section. This is valid when the sampler
is immersed in a relatively large cloud of radiative material, and the dominant cloudshine
contribution is determined by the local concentration. For situations where the cloud
remains elevated, producing little surface dose or deposition, and radiative cloudshine
from remote material is dominant, the accurate method described below should be used
The formal definition of the gamma dose rate from a volume distribution of
radiating material is

Rcloud ( x ) = d 3 x d F ( x x , ) c ( x )

110

(8.28)

DRAFT
where the volume integral is over all space and denotes the radiation wavelength. The
factor, F, includes the energy of the emission and the transmission/absorption effects
along the path between the receptor and the emitter locations. The relation (8.28) must be
summed over all nuclides present in the material, since different nuclides have different
emission spectra and transmission properties. The gamma spectra and summation over
nuclides is pre-computed by the NFAC source module, and a combined effect is specified
so that only the spatial integral is required. Thus,
Rcloud ( x, t ) = d 3 x Fcloud

x x , t ) c ( x)

(8.29)

The function, Fcloud , is generated by the NFAC module, and accounts for the decay of the
material so that the function depends on time after release.
Direct calculation of (8.29) is still a demanding exercise since the local
concentration, c, is obtained as the sum over all the puffs in the calculation. We therefore
require a three-dimensional spatial integral for each puff, for each surface location, for
each timestep. Since the weighting function, Fcloud , is specified numerically, analytic
approximations are precluded and a numerical estimate of the spatial integral must be
used.
The efficient calculation of the cloudshine dose integral is accomplished by
1)

computing the dose rate only once per large timestep, tL

2)

computing dose rate on a relatively coarse surface grid

3)

using a coarse grid for the spatial integration

4)

restricting the range of the calculation for each puff

Details of the cloudshine computation are given in a separate report (Sykes et al., 1998).

111

DRAFT
SECTION 9
MODEL OUTPUT
9.1 LOCAL CONCENTRATION VALUES.
The description of the concentration field is based on summing contributions from
each puff using the generalized Gaussian shape given in (2.1). The only complicating
factor in this process is the treatment of reflections at the surface and possibly at the
boundary layer capping inversion. The notation is simplified by using D = Det(), and
using the tensor defined as

ij =

1
ij

(9.1)

Using these variables, the mean concentration field of an individual puff is

a f a2 f Q D

c x =

3/ 2

1/ 2

exp ij x i x i x j x j

gc

(9.2)

The usual reflection condition corresponds to a zero normal gradient boundary

condition on c, which is consistent with zero flux through the surface. In general, there is
a deposition flux at the ground but we assume that the flux is small so that a zero gradient
is justified. When the Gaussian is skewed in the vertical, however, there is no simple
reflection condition that maintains a zero normal gradient. The skewness effects are most
pronounced under stable conditions with vertical wind shear, and are generally small
under the diffusive conditions of the turbulent surface layer. The skewness can be
important under stable conditions with terrain, where horizontal impact on a hillside can
occur. We therefore require a simple self-consistent reflection condition for skewed
puffs.
For a surface reflection, the location of the maximum surface concentration in the
unreflected Gaussian (9.2) is obtained by examining the exponential argument at z = h.
Here, h is the local terrain elevation, and if the terrain slopes are non-zero then the puff
moments are all rotated into the local coordinate frame one axis normal to the surface
prior to the reflection calculation. The surface location, x s , y s , is given by

112

DRAFT
x s x = 13 22 23 12

g Dz

(9.3)

g Dz

(9.4)

y s y = 23 11 13 12

where
2
DH = 11 22 12

(9.5)

The reflected puff is then positioned on the projection from the original puff
centroid through the surface maximum location, i.e.

xr = 2xs x

(9.6)

where x s = x s , y s ,0 . The reflected puff moments are the same as the original puff, so
that the skewness is the same. For a non-skewed puff, the reflection is the normal vertical
reflection but a skewed puff will reflect from the surface maximum point back toward the
original centroid. For the special case where z = 0, the reflection will simply double the
concentrations in z > 0. This reflection does not give a zero slope at z = 0 for a skewed
puff, but is the simplest mass-conserving scheme that maintains a smooth transition as a
puff approaches the surface.

The total concentration in z > 0 can be written as the sum of the original and
reflected puffs, giving

a f a f a2 f Q D

cT x = c x +

3/ 2

1/ 2

exp ij x i x ri x j x rj

gc

(9.7)

where c is defined by (2.6). Using the definition of the reflected centroid in (2.11), (2.13)
can be rewritten as
c T x = 1+ exp - Rz c x

af b

a fg a f

(9.8)

where the reflection coefficient, R, is given by

R=

z
2 DDH

(9.9)

For puffs located within the planetary boundary layer, i.e., z < z i , the puff mass is
assumed to be entirely contained within the turbulent layer. This is consistent with the
assumption that vertical diffusion across the capping inversion is small. Reflection at the

113

DRAFT
inversion, z = zi, is treated similarly to the surface reflection and gives an additional
reflection factor of

f r = 1+ exp - R zi z

(9.10)

for points within the boundary layer, where

R=

zi z
2 DDH

(9.11)

If the puff centroid is within the boundary layer, then c = 0 for z z i .

9.2 PROBABILISTIC CALCULATION.

9.2.1 Probability Density Function.

The SCIPUFF prediction does not provide a complete description of the

probability distribution for the concentration value, but the two moments (mean and
variance) can be used to generate an empirical distribution function with a specified
shape. Analysis of atmospheric observations (Lewellen and Sykes, 1986; Yee, 1990) has
shown that the truncated Gaussian or clipped normal distribution provides the best
description of short-term concentration fluctuations if only the mean and variance are
available. The clipped normal distribution is obtained by replacing all the non-physical
negative values in a general Gaussian distribution with zero values. Thus a Gaussian
distribution

p G (c ) =

F b c g I
expG
2
H 2 JK
2

G
2
G

(9.12)

with mean G , and standard deviation G , becomes

2
2 G

exp

G 2

( c G )

pC (c) = ( c ) +

, c0

(9.13)

where the intermittency, , is given by

G
1
1 erf
2
G 2




114

(9.14)

DRAFT
The mean and variance of the clipped distribution, pC , can be related to the
Gaussian parameters.

" $


  

G 2

 %!! #

2
2 = 2 + G 1 + erf

1 + erf

and

G
G 2

 

2
G
G
exp
2
2 G
2

+ G

(9.15)

(9.16)

The cumulative distribution function, CDF, is a more useful quantity than the
probability density function, pC . The CDF is normally defined as the probability that the
random variable (concentration, or dosage, for example) is less than or equal to any given
value, C, i.e.,
F ( C ) = Pr ( c C )
C

(9.17)

&

= pC ( c ) dc
0

22 (844 06
22 (844 06

)335 '117

Thus, for the clipped normal distribution, (9.13), we have

C G
1
1 + erf
2
G 2

)335 '117

F (C ) =

,C 0

(9.18)

Also, the probability that c exceeds any value, C, is simply 1F, i.e.,
Prob ( c C ) =

C G
1
1 erf
2
G 2

,C 0

(9.19)

9.3 POPULATION/AREA EXPOSURE.

SCIPUFF gives a probabilistic prediction of the concentration or dose, c, as a
function of spatial position on the ground. In order to assess population exposure, the
field can be integrated over the surface area and multiplied by a population density to
provide a simple estimate of the number of people exposed to a given dose or
concentration level. The population exposure assessment is activated by checking the
"Population Estimate" box in the plot control panel and choosing color area fill for the

115

DRAFT
contour plot display. A simple calculation of the spatial area within each display contour
is available if "Area Estimate" is checked instead.
The population density, , can be selected either from a worldwide database or as
a single number that applies to the entire plotted domain. If a uniform density, , is
selected, the units are persons per square kilometer, and should be chosen to represent the
area of interest in the plot region. Two simple "Population Estimate" measures are
provided in the plot output window, depending upon the type of plot selected.
If the mean dose/concentration is plotted with contour levels {Xi , i=1, N}, then a
mean exposure is calculated for each contour level from the following integral:

N i = Pr ( c X i ) dA

(9.20)

where the probability of exceeding Xi is calculated from the clipped normal assumption, as
in (9.19). Thus, Ni is the average number of people exposed to a dose/concentration level
above Xi, and is displayed in red above the appropriate contour value in the color table to
the right of the plot. Note that "Area Estimates" are displayed in blue and are always a
simple area calculation for the displayed contour levels.
If the probability of exceeding a specified level, X, is plotted with contour levels
{pi, i=1, N}, then the number of people exposed at a risk of pi, is calculated from the
following integral:
Mi = I ( x ) dA

(9.21)

where the indicator function I marks the area where the probability is greater than or equal
to pi, and is defined as
I=

R1 if Pra c > X f p
S

T 0 otherwise

(9.22)

Thus in this case, Mi represents the number of people exposed to a level of X with
risk pi, or greater and is displayed in red above the appropriate contour value in the color
table to the right of the plot.
Care should be taken in interpreting these probabilistic population estimates. The
two definitions given above for N and M are precise; they represent the average number

116

DRAFT
of people exposed to a given level and the number of people exposed at a given risk level,
respectively. However, the average number, N, may not be the most likely number, and
the number of people, M, with risk p does not necessarily imply a risk p of exposing M
people. The probability distribution for the number of people exposed depends in a
complicated way on the spatial distribution of both the population and the
dose/concentration field. For example, a very uncertain plume direction will give a low
risk for any particular person if the plume is narrow compared with the direction
uncertainty. This will produce a large region of exposure at a low risk. However, if the
population is uniformly distributed over the region, there is a very high probability of
exposing some people since every realization of the plume will produce a surface impact
somewhere assuming vertically well-mixed conditions. In this situation, the most likely
number of exposures will be close to the average exposure, N. Since any realization is a
localized plume, there is very small risk of exposing all M people, but also a very small
chance of exposing no people.
For inhomogeneous population distributions, such as localized urban centers, the
situation is more complicated and the probability distribution is wider. Suppose the
population is all located in a relatively small region compared with the size of a typical
area of impact, so that the plume essentially either hits (completely covers) or misses the
urban region. In this case, there is a finite chance, p, of exposing the entire population,
M. However, there is also a significant possibility, 1 p, that a realization will miss the
urban center and produce a zero exposure.

9.4 HAZARD AREAS.

Any prediction of dispersion in the atmosphere is uncertain to some degree. In
addition to the random nature of the turbulent diffusion process, there are many other
sources of uncertainty in a dispersion calculation. One of the major sources is the
meteorological input, which may be based on interpolation or extrapolation of
observations or could be a numerical forecast field. Other causes of uncertainty include
incomplete knowledge of the source, inadequate modeling of the physics (dispersion,
deposition, etc.), and numerical inaccuracies. An assessment of the uncertainty in a
model prediction is an essential part of any planning/response application, since it must
be recognized that a deterministic "correct" result is impossible for real situations. The

117

DRAFT
uncertainties are highly dependent on the available information, as well as the type of
output required, and it is important to provide some indication of reliability.
A proper treatment of the entire uncertainty is beyond out current capability, but
SCIPUFF provides the framework to address some of the important aspects. Specifically,
the uncertainty in the wind field can be considered as a large-scale horizontal turbulence
field. For short-range dispersion this can be represented as a wind direction uncertainty
under some conditions, but is more generally described statistically in the same fashion as
the boundary layer turbulence. The prediction of hazard areas is based on a dispersion
calculation that includes the large-scale velocity variances representative of the
uncertainties in the input wind field. The estimation of the wind-field uncertainties as a
function of the meteorological input is described in Section 12.4.
When the HAZARD calculation option is selected, a plot of "hazard area" is
available from the output interface. The output field is intended to provide an estimate of
the hazard region, and is based on a conditional probability calculation. The conditional
probability is simply obtained from (9.13) as the non-zero portion of the clipped normal
distribution, i.e.,
pC (c ) =

a 1 f

F b c
expG
2
H 2

G
2
G

g IJ ,
K
2

c0

(9.23)

The conditional probability ignores the intermittency in the distribution, and gives the
probability of exceeding a threshold value conditional upon experiencing a non-zero
concentration or dosage.

9.5 CASUALTY OUTPUT.

9.5.1 Prompt Nuclear Weapon Effects

Prompt weapon effects, representing combined effects of nuclear blast, thermal

radiation and prompt gamma radiation, are based on output from PDCALC (Wright,
1996). This routine provides a radial distribution for the probability of fatality, Pf i r ; k ,
and the probability of casualty, Pc i r ; k , as a function of the individual weapon
characteristics and also the population protection posture. The specific weapon is

118

a f

DRAFT
denoted by the index i, and the protection posture is denoted by the index k. The radial
distributions are used to generate circular field values, and are accumulated on an
adaptive grid (see Section 3.7) to assess the total prompt effects.
Assuming the independence of individual weapons gives the probability of
surviving N events as the product of the individual probabilities of survival, i.e., the
complement of the probability of fatality. Thus, we accumulate

ln 1 Pf i

ax ; k f =

x x 0i ; k

prompt

Ff

jj

(9.24)

where x0i denotes the location of weapon i. The final probability of prompt fatality due
to the collection of weapons is then given by

a f

a fj

Pf x ; k = 1 exp Ff prompt x ; k

(9.25)

a f

A similar expression gives the probability of casualty, Pc x ; k . Note that fields

are stored for each protection class, k, for later use in conjunction with the population
database.

Note that the prompt effects are computed for the complete list of weapons, and
are not stored as a function of time. Thus, when a number of weapon sources are
specified, the output display is not able to present prompt effects at different times during
the calculation; only the total is available.

9.5.2 Fallout Radiation Effects

As discussed in Section 8.2, SCIPUFF integrates the external radiation dose due
to radioactive deposition from a nuclear weapon on the surface grid. This information is
used to estimate fatality and casualty due to longer-term exposure to radiation. The
human effects are described by the response function, Ff( fallout ) D, R, trel , which gives the
probability of fatality as a function of the exposure. Here D is the integrated dose, R is
the dose rate, and trel is the effective release time for the weapon. These three quantities
are used to provide an estimate of the exposure history. The response functions for both
fatality and casualty are based on the RIPD algorithms (Anno et al., 1996). The fallout
casualty algorithms also account for protection posture by applying a reduction factor to
the dose and dose rate based on the protection category, k.

119

DRAFT
Since SCIPUFF provides both the mean dose, D , and also the standard deviation,
D , the mean response is estimated by integrating over the predicting probability
distribution for the radiation dose. Thus,

( x; k ) =

( fallout )
f

p ( ) Ff( fallout ) ( k D, k R , trel ) d

(9.26)

where p is a clipped normal distribution with unit mean and a standard deviation of
D D , and the mean dose parameters are evaluated at the spatial position, x. Here, k is
the protection factor, which is used to reduce the exposure in a simple linear fashion. A
similar calculation is performed for the mean casualty response, Fc( fallout ) .
9.5.3 Casualty Estimates

Casualty estimates are computed for nuclear weapon scenarios, and include both
prompt nuclear effects and radiation fallout effects. Several measures can be computed
and the HPAC output tables contain three different estimates, labeled 'Best Estimate',
'Worst Case', and 'At Risk'. All three measures use a worldwide population database,
which provides both total population numbers for arbitrary regions and also a count by
protection class. The different protection classes are used to mitigate both prompt and
fallout radiation effects, and the user can choose to use or ignore these effects in the
casualty table calculations.
The tables are all constructed as a 33 matrix as follows:
Prompt Fatalities, N (f prompt )

Fallout Fatalities, N (f fallout )

Total Fatalities, N (f total )

Prompt Injuries, Ni( prompt )

Fallout Injuries, Ni( fallout )

Total Injuries, Ni( total )

Prompt Casualties, Nc( prompt )

Fallout Casualties, Nc( fallout )

Total Casualties, Nc( total )

where casualties represent the sum of fatalities and injuries, and the totals represent the
sum of prompt and fallout radiation effects.
9.5.3.1 Best Estimate Table
The "Best Estimate" table is computed as the expected or mean value for each of
the 9 table entries. The prompt fatalities are computed first, as
120

DRAFT
Pf x ; k pop x ; k

N (f prompt ) = dA

a f

a f

(9.27)

where pop x ; k is the population density in protection class-k at the grid location, x. The

a f

integral is taken over the area of the calculation domain, and is actually calculated
numerically on the adaptive SCIPUFF grid. This assumes that the prompt effects occur
first, even though the weapon releases may be specified at different times during the
calculation.
The fallout radiation fatalities are computed next on the population remaining
after prompt fatalities have been discounted. Thus,

c 1 P ax ; k f h F

N (f fallout ) = dA

( fallout )
f

ax ; k f ax ; k f
pop

(9.28)

Total fatalities are obtained as the sum of the prompt and fallout contributions,
i.e.,

N (f total ) = N (f prompt ) + N (f fallout )

(9.29)

The next table entry that can be simply defined is total casualties, since this
represents the complement of the set of all "unharmed" people. We do not account for
combined effects of fallout radiation and prompt weapon effects; the two are treated
independently. Thus the probability of being a casualty from either effect is

Pc( total ) = 1 1 Pc( prompt ) 1 Fc( fallout )

(9.30)

The total casualties are then obtained as

Pc( total ) x ; k pop x ; k

a f

Nc( total ) = dA

a f

(9.31)

The partition of the total casualties between prompt and fallout is not
unambiguous, but the method used here first estimates the prompt contribution directly,
then corrects the estimate for fallout radiation effects. Thus, in the absence of fallout
radiation, the prompt casualties would be calculated similarly to the fatalities in (9.27).
However, a fraction of these casualties will be counted amongst the fatalities attributed to
fallout radiation. The correction term therefore multiplies the non-fatal prompt
casualties, i.e. prompt injuries, by the probability of fatality due to fallout radiation.
Thus,

121

DRAFT

Nc( prompt ) = dA

cP

( prompt )
c

Ff( fallout ) Pi ( prompt ) pop x ; k

a f

(9.32)

where Pi ( prompt ) = Pc( prompt ) Pf( prompt ) is the probability of prompt injury.
The remaining table entries are simply constructed from the addition rules, i.e.,
total is the sum of prompt and fallout, and casualties are the sum of fatalities and injuries.
9.5.3.2 Worst Case Table
The "Worst Case" table is computed in a completely analogous manner to the
"Best Estimate" but use conditional probabilities for the radiation dose distributions
instead of the absolute probability. Note that for "DUAL" runs, the results with
meteorological uncertainty included are used for this table. The prompt effects are not
currently implemented with conditional probabilities, so they are identical to the "Best
Estimate". However, the fallout radiation dose probability distribution is assumed to be
clipped normal, and therefore the distribution p in (9.26) is associated with a finite
probability, , of non-zero value as described in Section 9.2.1. The casualty estimates for
fallout radiation are therefore scaled by the probability of a non-zero dose for the "Worst
Case" table, so that the fallout fatalities, for example, become

af

= dA

c 1 P ax ; k f h

( fallout )
f

a f ax ; k f

Ff( fallout ) x ; k

pop

(9.33)

All other table entries are constructed in the same way as in the "Best Estimate"
table, but fallout probabilities are divided by as defined by the radiation dose
distribution.
9.5.3.3 At Risk Table
The third output table represents the population exposed to a prescribed level of
risk, r0, which is based on the conditional probabilities for the fallout contribution. Thus,
population at risk to prompt fatality is given by
H Pf x ; k r0 pop x ; k

c a f h

N (f prompt ) = dA

a f

(9.34)

where H is the Heaviside step function, which is unity when the argument is positive and
zero otherwise. Similarly, the total number at risk of fatality is

122

DRAFT
H Pf( total ) x ; k r0 pop x ; k

N (f total ) = dA

a f h

a f

(9.35)

where

Ff( fallout )

(9.36)

Pf( total ) = 1 1 Pf( prompt ) 1

similar to (9.30). The number at risk of fallout fatality is defined as the difference
between the total and prompt, as in (9.29).
Numbers at risk of casualty are computed in the same way as fatality, but using
casualty probabilities instead. Thus, for example,

H Pc( total ) x ; k r0 pop x ; k

Nc( total ) = dA

a f h

a f

(9.37)

where

h FGH

Pc( total ) = 1 1 Pc( prompt ) 1

IJ

Fc( fallout )

(9.38)

Finally, numbers at risk of injury are calculated similarly using probability of

injury, which is defined as the difference between probability of casualty and probability
of fatality. The injury probability is defined separately for prompt and fallout radiation
effects, and combined to give a total probability of injury as in (9.38).
Note the numbers in the "At Risk" table sum horizontally, so that the total is the
sum of prompt and radiation fallout contributions. However, the casualty estimate is not
the sum of fatality risk and injury risk, since population can be counted as being at risk
for multiple effects.

123

DRAFT
SECTION 10
METEOROLOGY SPECIFICATION
10.1 BACKGROUND.
The advection and diffusion processes modeled with SCIPUFF require
meteorological input of some kind. The minimum requirement is a single wind vector.
Obviously, more meteorological input is usually desirable and SCIPUFF can assimilate
observational data ranging from a single wind measurement to multiple profiles which
include turbulence measurements and/or boundary layer parameters such as mixing layer
height and Pasquill-Gifford-Turner stability class. Additionally, with the specification of
topography, SCIPUFF will calculate an adjustment to the three-dimensional velocity field
obtained from interpolating observations so that it has very low divergence, i.e.,
conserves mass. Alternatively, three-dimensional gridded wind and temperature fields
generated by some prognostic model or analysis may be input.
SCIPUFF meteorological input is specified as one of two generic types:
observational or gridded. Observational data is characterized by irregularly spaced
vertical profiles with irregular horizontal locations, as well as irregular reporting times.
The number and locations of observations can vary from one time to the next.
Observational input is specified through a file of with vertical profiles and/or a file of
single level surface data. Gridded meteorological input is defined on a regularly spaced
three-dimensional grid; time need not be regularly spaced. The gridded input option is
intended for use with numerically generated wind fields.
The formats and other requirements for the two types of meteorological input files
are given in Section 13.2

10.2 OBSERVATIONAL INPUT.

Observation input can be of two types. A SURFACE observation file contains
multiple records of the mean wind, temperature, and possibly boundary layer parameters.
124

DRAFT
Each record represents a station location at a specified time, and the file can contain
multiple stations and multiple times stored in chronological order. The surface
observations are assumed to be given at the reference height, zref, unless the height is
provided as part of the observation record. The reference height can be specified on the
surface file or is otherwise assumed to be 10m. However, if the heights are given on the
observation record, zref is redefined as the mean observation height at the nearest time
before the start of the run. A PROFILE observation file is similar in character to the
surface file and in fact it can optionally contain surface observations such as heat flux and
mixing-layer height. (See Section 13.2.1 for a list of acceptable variables). For a given
station, there are typically multiple levels of observations, although a single observation
level can be specified.
The meteorological observations are interpolated onto a rectangular grid for use in
SCIPUFF, so that local values can be rapidly found for arbitrary puff locations. When
the mass-consistent adjustment option is invoked, the vertical grid must be input.
Otherwise the vertical grid is defined from the observational input. If PROFILE data are
given, the vertical grid is based on the heights from the station that has the most levels
with valid wind measurements up to Zmax (the puff calculation domain height). For
Observation type files (the default type; see Section 13.2.1), this station is selected from
the first 30 observations on the file, but may be modified if a more appropriate station is
found at the nearest time before the start of the run. Since the vertical grid for Analysis
or Forecast type files is assumed to be constant, the grid is taken from the first
observation on the file. Finally, the grid is checked and modified if necessary so that it
has at least one level near zref suitable for SURFACE data if given. However, if only
SURFACE data are available, only one vertical level at zref is used.
The horizontal grid is initially assumed to be a single point but is expanded if and
when the number of observation stations is two or more during the SCIPUFF calculation.
A uniform horizontal grid is then constructed based on the puff domain boundaries, xmin ,
xmax , ymin , ymax and horizontal resolution H. It is assumed for the following discussion
that the horizontal grid has more than a single point.

The assimilation of observational data into the three-dimensional gridded wind

field is based on a weighted interpolation scheme. The weights given an observation are
functions of the horizontal and vertical distances, temporal separation and terrain
elevation relative to the grid points. In the following, it is assumed that both SURFACE
125

DRAFT
and PROFILE observations are available, but the methodology is also valid if either set is
missing. The terms corresponding to the missing data type are simply ignored. Data
assimilation is performed at times for which either SURFACE or PROFILE observations
are available. At such a time, the interpolation scheme for any point on the grid gives

i =1
NS

wo +

w +
S
i

i =1

S
i

NP

wiP iP

i =1

w +
S
i

(10.1)

NP

NS

n =

w o

n 1

w
i =1

P
i

where n represents any variable such as u, v, T, etc. at time t n , n 1 is the interpolated

value from the previous time t n1 , iS is the observation at the ith SURFACE station at time
t S , wiS is its corresponding weight, iP is vertically interpolated from observations at the ith
PROFILE station at time t P , wiP is its corresponding weight, N S and N P are the number
of valid surface and profile observations respectively at t S and t P , and wo is a weight
assigned to the field at t n 1 . Note that either t n = t S or t n = t P , so that t S , t P t n . Also
note that the weights can vary for different field variables (but for visual clarity this
dependence is not indicated in the notation). The definition of the weights, which depends
somewhat on whether an observation is SURFACE or PROFILE, will now be discussed.

In general, an observation weight is determined as the product of three functions

dependent on the horizontal, vertical and temporal separations between the observation
and grid point:
w obs = wh (r ) wv ( z ) wt (tobs tn )

(10.2)

where w obs represents wiS or wiP , r is the horizontal distance from the observation location
to the grid point, z is a measure of the vertical separation (accounting for terrain
differences) and t obs is t S or t P as appropriate.
The horizontal weighting function is simply based on the inverse distance
squared:

wh = 1 + r 2 a 2

(10.3)

where the core radius a is used to avoid the singularity as r approaches zero and reduces
the complete dominance of an observation near the grid point; a 2 = 0.25 x 2 + y 2
where x and y are the horizontal grid spacings.

126

The vertical weighting function has a similar form:

DRAFT
w v = 1 + C z 2 a 2

(10.4)

where C is a constant set to 1000. The purpose of setting C large is to give extra weight to
observations which are close vertically in preference to observations which may be close
horizontally but relatively distant vertically. Although z will defined shortly, for the
time being assume that it is simply the vertical separation (which is the case over flat
terrain). Then, an observation with z 0 but r = 0 is given the same weight as one with
no vertical separation but r 32 z (assuming C z 2 >> a 2 ). This is motivated primarily
by the vertical stability of the atmosphere, which tends to reduce velocity correlations as
vertical separation increases. The following expression for z is an ad hoc attempt to
account not just for vertical separation but also for terrain differences:

z = z +

3
h

(10.5)

2
2
h + zh +

where h = h hobs , h is the terrain height at the grid position, hobs is the terrain height at
the observation location, z h = max z h, ~ hobs , z is the grid height (above the reference
z
z
z
or zero height level), ~ = max z, z h + hobs , z = ~ zobs and z obs is the height of the
z
nearest observation level used to interpolate linearly to ~ . is a small number (set to
10
10 ) which avoids division by zero when h = z h = 0 .

As indicated above, iP is found from a linear interpolation of the profile data at

z
z
the height ~ . However, if ~ is below the lowest observation level, iP is set to the value
at the lowest level for temperature and humidity while horizontal velocity components
are reduced using a surface layer profile, including stability effects (see Section 11). If ~
z
is above the highest observation level, iP is set to the value at that level, except for
temperature which is extrapolated using the temperature gradient from the U.S. Standard
Atmosphere.
For surface observations, iS is assumed constant with height for temperature and
humidity while velocity components are scaled using a surface layer profile. Thus, the
velocity components are reduced when z < zobs and increased for z > zobs , up to the
surface layer height. However, the vertical distance, z , used for calculating wv is still
based on the height of the observation.
When combining SURFACE and PROFILE temperature observation, the vertical
weight, wv , is modified for the SURFACE observation: if the height of the grid point

127

DRAFT
above ground is greater than 500m, wv is set to zero; otherwise, the expression from
(10.4) is multiplied by a factor min 500 ( z h) 300 ,1 . Thus, the temperature weight
is unchanged for z h 200 m , but goes linearly to zero at 500m. This factor is used
since a surface temperature measurement will usually not be representative of the
atmosphere well away from the surface, particularly above a mixed layer or under stable
conditions.

When only SURFACE temperature observation are available, the U.S. Standard
Atmosphere profile is used, shifted so as to match the observations at the surface.
The temporal weight uses a simple exponential aging function:

wt = exp( tn tobs obs )

(10.6)

The time constant obs is set to 6 hours if t n is beyond the last or ahead of the first
observation time; otherwise it is based on the time between observations and the number
of observations so that times with many observations are given relatively large weights.
Thus,
~
obs = f t obs N nobs N obs
(10.7)
1
where f is an empirical constant set to 0.4323, t obs is the time between the current and
~ obs
previous observations and N obs is N S or N P . N n 1 is a measure of the number of
observations from the previous time which accounts for time-smoothing and is dependent
of the history of the data assimilation. It is updated when t n = t obs using obs from (10.7)
and the current number of observations:
~ obs
~ obs
N n = N obs + N n 1 exp t obs obs
(10.8)
~ obs
~ obs
where initially N 0 is set to zero. As an example, if N obs and t obs are fixed, then N n

.
rapidly approaches an asymptotic value of 116 N obs .

Temporal smoothing is chiefly implemented though the weight wo given to the

fields at t n1 . It is a function of the time between updates, t n = t n t n 1 , and the
observation weights used at t n1 . For a given grid location, a time-smoothed sum of

SURFACE observation weights is defined as

S
n 1

SnS =

S| S

S
n 1

exp t n S

NS

wiS

RS
|

t S > tn

tS = tn

i =1

128

(10.9)

DRAFT
where S is the time constant from (10.7) appropriate to SURFACE observations and
S0S = 0 . The sum for PROFILE observations, SnP , is defined similarly. Then,

w o = SnS1 + SnP1

(10.10)

This definition is an attempt to quantify the relative quality of the interpolated threedimensional fields, e.g., fields at t n1 which assimilated many observations will dominate
over a single observation at t n as long as they are not too old.

10.3 GRIDDED INPUT.

10.3.1 File Format Types.

SCIPUFF supports two types of gridded formats designated as MEDOC and

HPAC. The MEDOC format is based on the French Electricity Board pollutant
dispersion programs as developed for DSWA application by SAIC under contract number
DNA001-92-C-0151. The HPAC format was defined by SAIC (McGahan, private
communication) for interfacing with the National Weather Service Medium Range
Forecasts. Both formats are described in Section 13.2. The HPAC format is limited to
latitude/longitude horizontal grids and no terrain input is available. Gridded data in other
formats must first be converted to one of these two input formats.

10.3.2 Terrain.

The MEDOC format allows the input of terrain elevation information (but it is not
mandatory). It is then assumed that the coordinate system on the file, (, , ), is terrainfollowing as defined by the transformation of Clark (1977):

=x
=y
=

(10.11)

zh
J

where
J = 1

h
D

(10.12)

129

DRAFT
Here h(x,y) is the local terrain elevation, x and y are the Cartesian horizontal coordinates,
z is the Cartesian vertical coordinate, and D is the depth of the model domain above the
zero datum (typically mean sea level). It is assumed that the Cartesian velocity
components are on the file. Then, only spatial gradients need be transformed in SCIPUFF
as follows:

=
+ Gx
x

=
+ Gy
y

(10.13)

= J 1
z

where

represents any variable such as a velocity component or temperature and

F 1I
HD K
F 1I
=
HD K

Gx =

h
x

Gy

h
y

(10.14)

Simple finite differencing is employed to compute the partial derivatives in (10.13).

However, the vertical gradient of vertical velocity, w, is treated as a special case and is set
to insure satisfying continuity:

w
u v
=
+
z
x y

(10.15)

10.4 MEAN FIELD INTERPOLATION.

The mean meteorological fields are updated every large timestep tL using simple
linear interpolation. If ( t ) represents a field at time t, then the time interpolation is
defined by
t L
t L
( t t L ) +
(TM )
TM t + tL
TM t + t L

(t ) = 1

(10.16)

where TM is the next time when the fields are given. Note that storage for only two time
levels is required.

130

DRAFT
The computation of mean wind and temperature at a puff centroid can depend on
its proximity to the surface and its vertical position relative to the inversion (if one exists)
in conjunction with the meteorological input type and boundary layer specifications.
However, under all circumstances, horizontal interpolation is required (except for the
trivial case of a single location for the meteorology). Therefore, we now describe the
simple bilinear interpolation employed in SCIPUFF.
The background meteorology is always defined on a rectangular, uniformly
spaced horizontal grid so that x i = x1 + i 1 x , with yi defined similarly. The locations
of the four surrounding grid locations are x i + m , y j + n , m, n 0,1 where x , y is the
puff centroid, i p = int x x1 x + 1 and jp is defined similarly. Then, the value of any
field at the centroid location, represented by , is

i =0 j =0

where

fi x = 1 i + x x i

f jy = 1 j + y y j

fi x f jy

p +i, j p + j

a f

a f

(10.17)

a f
e

(10.18)

and i i , j j is the field variable at x i +i , y j + j . Where vertical interpolation is required

(all cases except SURFACE observations only), a similar procedure is used, i.e., (10.17)
will have a third summation with a vertical factor f kz .
p

The interpolated fields may be modified due to the presence of the planetary
boundary layer. For instance, if a puff is located below the lowest grid level, the velocity
must be scaled according to the surface layer profile. This is discussed further in Section
11.

10.5 MASS-CONSISTENCY.
A simplified three-dimensional mass-consistency scheme is available as part of
the SCIPUFF meteorological input. If terrain input is specified in conjunction with
observational wind input or certain kinds of gridded input, the interpolated wind field is
then adjusted to ensure mass conservation, i.e., its divergence is zero (or at least very

131

DRAFT
small). It should be noted that this feature is invoked only if the mass-consistent module
SWIFT is unavailable.
The adjustment procedure uses either an iterative Fast Fourier Transform (FFT)
solver or a point relaxation scheme, depending on the magnitude of the effective terrain
slopes. The FFT solver is faster but converges only for relatively small slopes. The
effective slope is defined as the actual slope, s, divided by the vertical partition
parameter, . The partition parameter depends on the vertical stability and is usually
small, resulting in near-horizontal flow around terrain features. The adjusted velocity
fields are calculated as

xi

ui( mc ) = ui(0) +

(10.19)

where ui(0) is the interpolated field based on the input winds, and satisfies the Poisson
equation
2 2
2
+ 2 + 2 = u(0)
x2 y
z

(10.20)

This equation is solved in a non-orthogonal terrain-following coordinate system,

but the departure from a Cartesian system is small for small slopes. In this case, the
direct FFT scheme is convergent. However, the slopes are too large in most cases, and a
time-like iteration based on the artificial compressibility method is used. For each
iteration, denoted by n, we define
u ( n ) = u ( n 1)
v

(n)

(n)

(n)

=v

( n 1)

( n 1)

( n 1)

=w

(
=

( n 1)
x

( n 1)

(10.21)

( n )

2 u( n 1) )

1 + k

and the pseudo-timestep, , is chosen locally to satisfy a Courant condition. Note that the
vertical velocity adjustment is made implicit by coupling it with the calculation of . This
helps avoid restrictive Courant conditions in the typical case where the vertical grid
resolution is much finer than the horizontal. The damping coefficient k = 0.15 was found
132

DRAFT
necessary to achieve numerical convergence. However, since approaches zero as the
velocity field becomes divergence-free, k should not affect the final solution.
The iterative scheme is continued until the maximum divergence in the domain
falls below the convergence criterion, Dmax , where the divergence scale, Dmax , is
(0)
(0)
max ( umax x, vmax y ) 10 and can be selected by the user. The default value for is
0.01.
For small slopes, the elliptic equation is cast as the sum of the standard separable
Cartesian operator plus non-orthogonal terrain terms.
2 2
2
+ 2 + 2 = u(0) + L ( )
x2 y
z

(10.22)

where the spatial coordinates are now the transformed variables, and L represents the
terrain terms. This equation is solved directly for ( n ) using ( n1) on the right hand side.
The iteration is continued until the maximum change in divided by the maximum value
of is less than the specified criterion.
The vertical partition factor, , is defined empirically as
1
1 + Fr 2

(10.23)

where the Froude number is given by Fr 2 N 2 x 2

y 2 u 2 v 2 and N 2 is the
Brunt-Vaisala frequency. It should be noted that = 1 corresponds to potential flow (the
flow tends to go over and around terrain) while = 0 results in flow constrained to a
horizontal plane (around terrain). The user can specify maximum and minimum values
for .

133

DRAFT
SECTION 11
PLANETARY BOUNDARY LAYER
11.1 GOVERNING PARAMETERS.
Defining the mean wind and turbulence profiles is essential in characterizing the
transport and turbulent diffusion rates in the planetary boundary layer (PBL). However,
since it is unlikely that detailed profiles will be available in routine use of SCIPUFF, it is
necessary to consider some idealizations of the PBL structure. Much research on the
PBL has been undertaken over the last forty years, with the result that there is a general
consensus on its overall structure, the applicability of similarity laws and the governing
parameters, at least for stationary flow over flat, homogeneous surfaces. (See Wyngaard,
1985; Venkatram and Wyngaard, 1988; or Lewellen, 1981.) It is now recognized that
under ideal conditions both the mean wind and turbulence are functions principally of the
following parameters:
1)

surface roughness length, z0,

2)

boundary layer depth, zi,

3)

Monin-Obukhov length, L, and

4)

surface friction velocity, u .

Except for the boundary layer depth and possibly the roughness length, these parameters
are not measurable quantities. Therefore, they must be related to other quantities that may
be directly measured or at least estimated sensibly. So, instead of friction velocity and
Monin-Obukhov length, SCIPUFF utilizes a reference velocity, uref = (uref,vref), defined at
a height, zref, and the sensible heat flux, H. The various ways of specifying the input
parameters will be discussed in the following section, but first we relate them to the PBL
parameters L and u as well as the temperature scale, ,and the convective velocity scale,
w , which is used to characterize turbulence under convective conditions. (Two other
important parameters, Ug, the geostrophic wind and f, the Coriolis parameter, are not
considered here since they determine the "turning" of the wind over the boundary layer
depth due to the earth's rotation. This effect is not incorporated into SCIPUFF unless it is
explicitly provided in the upper air wind fields.)

134

DRAFT
11.1.1 Monin-Obukhov length, L.

The Monin-Obukhov length is defined as

L=

u* 3 T a c p

(11.1)

gH

where is von Krmn's constant, T is the air temperature (either the actual absolute
temperature if available or more typically the base temperature T0 used for the Boussinesq
approximation), a is the air density, and cp is the specific heat at constant pressure. L
represents the ratio of the shear-driven turbulence production to buoyant production and as
such is a measure of the flow stability. Unstable convective conditions (positive heat flux)
are associated with a negative L while stable flows (negative heat flux) are associated with
a positive L. For neutral flow, L becomes infinite.

11.1.2 Friction velocity u .

The surface friction velocity, u , is defined by

u2

(11.2)

where s is the average surface stress. In most cases, the surface stress will not be
available from meteorological input data, so the surface friction velocity is determined
from the well-known logarithmic velocity profile law, corrected for stability effects (see
for example Businger, 1973):

u =

Uref

(11.3)

ln zref z0 + 1 m

where Uref is the reference speed and m is the stability correction term.
determined from
2
2
2
U ref = uref + v ref + u L2 + v L2

Uref is
(11.4)

where u L2 and v L2 are the horizontal large-scale variances, all given at the reference

height. The large-scale variances are included in the definition of the reference speed
since they represent wind velocity fluctuations on time scales longer than the boundary
layer response time and therefore contribute to the average surface friction velocity. (The
reference height, zref, used in (11.3) will, under certain circumstances, be the "surface
layer height" defined in Section 11.3; zref is required to be less than or equal to the surface
layer height in order for the surface layer relation (11.3) to apply.)

135

DRAFT
For neutral flow ( H = 0), m = 0. For the unstable case ( L < 0), m is given by

L c 1 + h OP
= 2 ln M
NM 2 QP
1
m

L c 1 + h OP
+ ln M
NM 2 QP
2
m

2 tan 1 m1 +

c h

(11.5)

where the non-dimensional velocity gradient m is given by

m = 1 15 z L

1 4

(11.6)

The most commonly used form for m under stable conditions ( L > 0), is (e.g.,
Businger et al. 1971; Dyer 1974; Hgstrm 1987)
m = Am z L

(11.7)

where Am has a value around 5. However, as Carson and Richards (1978) and others point
out, this form is valid only for z < L . They suggest that the Wangara data presented by
Hicks (1976) are suitable for describing m for z > L ; and this is confirmed by Holtslags
(1984) analysis of data from Cabauw. Van Ulden and Holtslag (1985) give an empirical
function which matches (11.7) for z < L and is also a good fit to the observations at larger
z:
m = 17 1 exp 0.29 z L

(11.8)

Due to uncertainties of the observations, there is some question as to the behavior of m

at large values of z L (above 5 to 10). However, as described in Section 11.3, (11.8) (and
(11.3) in general) is applied only within the surface layer and, under stable conditions,
the surface layer is limited to a maximum height of 5L.
The non-dimensional velocity gradient, as used in (11.6), is defined as
m =

z u

(11.9)

u z

For stable conditions, using (11.8) in (11.3) gives

m = 1 + 4.93 z L exp 0.29 z L

a f a

(11.10)

The calculation of u requires an iterative procedure since u is a function of L,

which in turn is a function of u . Also, the calculation is modified under stable
conditions if a CALCULATED boundary layer mode is specified; see Section 11.2.

136

DRAFT
11.1.3 Temperature scale .

The temperature scale for stable conditions is defined as

* =

H
a c p u*

(11.11)

and it appears in the temperature profile equation given by

T T0 =

*
0.74 ln z / z0 + 1 h

(11.12)

where T0 is the temperature at z 0 . We make use of (11.12) only under stable conditions,
in which case h = 4.7 z L . It should be noted that L can be defined in terms of * and
u as follows,

L=

2
u* T
* g

(11.13)

11.1.4 Convective velocity scale w .

Deardorff (1970) introduced the convective velocity scale, w , defined as

w* =

FG g H z IJ

1/ 3

(11.14)

H T c K
a

as the appropriate scale for convection-dominated turbulence ( L < 0). It is related to L

and u by

u3
L
= *3
w*
zi

(11.15)

11.2 SPECIFYING PBL PARAMETERS.

Specifying z0 and uref is generally straightforward, but other input requirements
depend on the PBL calculation mode chosen. There are four PBL modes referred to as
SIMPLE, OBSERVATION, CALCULATED and OPERATIONAL. In the SIMPLE mode,
boundary layer height and surface heat flux are estimated using simple functional forms
with user-specified minimum and maximum values. The OBSERVATION mode can be
used if measurements (or estimates) of zi and H, or L or stability class (PGT) are

137

DRAFT
available on an observation meteorology input file. In the CALCULATED mode, H is
estimated from a surface energy balance model and an evolution equation is solved for zi.
The OPERATIONAL mode is intended to utilize whatever relevant input is given in the
meteorological files and is essentially reset to OBSERVATION or CALCULATED mode
accordingly. Thus, the OBSERVATION mode is invoked if zi and H or L or PGT are
given, the CALCULATED mode is invoked if none of these variables are given, and if
either zi or H is available, the other will be calculated as with the standard CALCULATED
mode.
The specification of z0 and uref applies to all three modes. The CALCULATED
mode requires additional inputs including cloud cover and surface characteristics, as
discussed below.

11.2.1 Roughness Height.

Roughness length, z0, is a parameter related to the height of the roughness

elements (i.e. trees, buildings, etc.). Suggested roughness heights from Saucier (1987)
are given in Table 11-1. SCIPUFF can use a single roughness height for the entire
domain and duration of the run, or a two-dimension grid of values if specified on the
terrain input file (see Section 13.2.4).

11.2.2 Reference Height and Velocity.

For the purpose of defining PBL profiles, zref is taken from the lowest level of the
vertical grid. The vertical grid is defined either directly through the gridded meteorology
input file as described in Sections 10.3, 13.2.2 and 13.2.3 or constructed from
observational profiles as described in Section 10.2. uref is then the velocity at the lowest
grid level after the interpolation procedure (Section 10.4) and mass-consistent adjustment
(Section 10.5), if appropriate, are completed.

138

DRAFT
Table 11-1.

Suggested values for surface roughness (Saucier, 1987).

z0 (m)

Surface Description

0.0005

Bare Ground, Sand Dunes, Water

0.001

Nearly barren with low growing vegetation

0.01

Grassland, Cropland, Wetlands

0.05

Grassland with scattered trees, Brushland, Scrub growth

0.1

Deciduous Forest, Villages, Forest Clearings

0.5

Mixed Forest, Towns, Cities

1.0

Coniferous Forest

11.2.3 Surface Heat Flux.

The surface heat flux may be specified in three ways, depending on the PBL
mode selected:
1)

OBSERVATION - through the surface meteorology observation file

2)

SIMPLE - as a simple diurnally-varying function with prescribed daytime and

nighttime values

3)

CALCULATED - using a surface energy balance model which accounts for solar
radiation, cloud cover and surface characteristics.

When observation type meteorology is given and the OBSERVATION PBL mode
is specified, the surface heat flux can be specified directly in the observations surface file
(in Wm2) or it can be calculated from the Pasquill-Gifford-Turner (PGT) stability class.
The stability class is indicated on the observations surface file with a stability index. An
average Monin-Obukhov length is assumed within each stability class according to the
scheme described in Sykes and Lewellen (1992). The stability index and the
corresponding PGT stability class and Monin-Obukhov length are given in Table 11-2.

139

DRAFT
Table 11-2. Relationship between stability index, PGT stability class, Monin-Obukhov
length, L, and an assumed boundary layer depth, z i , if not specified as input.
Stability Index

PGT Class

L (m)

z i (m)

-5

1000

-12.5

1000

-50

1000

-1000

1000

25

125

13

65

25

Stability classes A, B, and C correspond to unstable conditions, D to neutral

conditions, and E, F, and G to stable conditions. The heat flux is then calculated from
equation (11.1) as
H=

3
u* T a c p

(11.16)

gL

When the SIMPLE mode is specified, the surface heat flux is calculated as
H=

R| H b H
S|
T

g FH 2 t IK ,
24

Hc cos

6 td 18

(11.17)

otherwise

Hc

where Hc is the constant nighttime heat flux, Hd is the maximum daytime heat flux, and td
is the time of day in hours. Typical values of Hc are between zero and 60 Wm2. Hd can
vary widely depending on cloud cover, time of year, surface characteristics, and latitude.
For example, on an overcast spring day in northern Russia, Hd might be only 10 Wm2,
while it could be around 200 Wm2 on a summer day in the southwest U.S.
In CALCULATED mode, SCIPUFF computes the sensible heat flux using a
surface energy balance method similar to that used in METPRO (Paine, 1987). The
method differs for daytime and nighttime conditions.

140

DRAFT
11.2.3.1 Daytime Heat Flux Calculation An analysis of the energy budget at the earth's
surface balances the net incoming radiation with the total surface heat flux. This requires
the partitioning of the surface heat flux into sensible and latent components. A simple
parameterization for such a partition proposed by Holtslag and van Ulden (1983) is
H=

Lv E =

a1 f s + b R Gg

(11.18)

s
Rn G +
s+

(11.19)

s+

where LvE represents the latent heat flux due to surface evaporation, Rn is the net solar
radiation at the earth's surface, G is the soil heat flux, estimated as 0.1Rn, is an
empirical surface moisture parameter defined below and is related to the Bowen ratio,

B = H Lv E , is an empirical constant set to 20 Wm2, s is the slope of the saturated

vapor-pressure curve and = c p Lv . Here, cp is the specific heat of air at constant
pressure and Lv is the latent heat of water vaporization.
The net radiation is estimated from the total incoming radiation using the method
of Holtslag and van Ulden (1983),

Rn =

SB T 4 + c2 nc
1 + c3

a1 rf R + c T
1

(11.20)

where R is the total incoming radiation, r is the surface albedo, nc is the fractional cloud
cover, c1 = 5.311013 Wm2 K6, c2 = 60 Wm2, c3 = 0.12 and SB is the StefanBoltzman constant, 5.67108 Wm2 K4.
The total incoming solar radiation is a function of solar angle () and cloud cover
and is computed from a formula suggested by Kasten and Czeplak (1980),

b
R = R0 1 + b1 nc

(11.21)

where R0, the incoming radiation for clear skies, is a function of and b1 and b2 are
empirical constants set to 0.75 and 3.4, respectively. A simple parameterization for R0
suggested by Holtslag and van Ulden (1983) is
R0 = a1 sin + a 2

(11.22)

where the empirical constants a1 and a2 are set to 990 Wm2 and 30 Wm2, respectively,
based on a survey of values for different northern mid-latitude sites.

141

DRAFT
The solar angle is determined from

sin = sin sin + cos cos cos h

(11.23)

where is the geographical latitude, is the declination and h is the hour angle (all in
radians). Given the approximate nature of these calculations, we follow Holtslag and van
Ulden (1983) in using simplified expression for and h, rather than the more accurate
formulas in Paltridge and Platt (1976), for example. Thus, is given by
sin = 0.398 sin s

(11.24)

where s, the solar longitude in radians, is evaluated from

s = 4.871 + d + 0.033 sin d

(11.25)

and d = d n 1 2 365.242 , dn being the Julian day number. The hour angle is given by

h = t 12 12 + + 0.043 sin(2 s ) 0.033 sin d

(11.26)

where t is universal time in hours, and is the Easterly longitude in radians..

Albedo is relatively constant for solar angles, , above 30, but increases for
angles below that. An empirical formula given by Paine (1987) is used to model the
variation of surface albedo with solar angle (in degrees):

r = r0 + 1 r0 e r + r

(11.27)

where r0 is the albedo for the sun directly overhead, r1 = 0.1 and r2 = 0.5 1 r0 .
As mentioned above, the surface moisture parameter a is related to B, the Bowen
ratio. Since B is a commonly used parameter, is expressed in terms of B using
equations (11.18) and (11.19) to give

s + Rn G

1 + B s Rn G + s +

(11.28)

This completes the description of the sensible heat flux computation in the
CALCULATED mode. It requires three more input parameters than the other two modes,
namely, fractional cloud cover, nc, surface albedo for the sun directly overhead, r0, and
Bowen ratio, B. The specification of cloud cover is self-evident. Tables 11-3 and 11-4
(from Paine, 1987) give suggested values for r0 and B as functions of land-use and
season.

142

DRAFT
Table 11-3. Suggested values for surface albedo as a function of land-use and season
(Paine, 1987).
Land-use Type

Spring

Summer

Autumn

Winter

Water

0.12

0.10

0.14

0.20

Deciduous Forest

0.12

0.12

0.12

0.50

Coniferous Forest

0.12

0.12

0.12

0.35

Swamp

0.12

0.14

0.16

0.30

Cultivated Land

0.14

0.20

0.18

0.60

Grassland

0.18

0.18

0.20

0.60

Urban

0.14

0.16

0.18

0.35

Desert Shrubland

0.30

0.28

0.28

0.45

Table 11-4. Suggested values for Bowen ratio as a function of land-use and season
(Paine, 1987).
Land-use Type

Spring

Summer

Autumn

Winter

Water

0.1

0.1

0.1

1.5

Deciduous Forest

0.7

0.3

1.0

1.5

Coniferous Forest

0.7

0.3

0.8

1.5

Swamp

0.1

0.1

0.1

1.5

Cultivated Land

0.3

0.5

0.7

1.5

Grassland

0.4

0.8

1.0

1.5

Urban

1.0

2.0

2.0

1.5

Desert Shrubland

3.0

4.0

6.0

6.0

143

DRAFT
In the CALCULATED mode, the heat flux may be used in computing the
boundary layer depth, as described in Section 11.2.4. It is also used to compute w* in
conjunction with zi using (11.14) and u* using (11.3). It should be noted that H could be
negative just after sunrise or before sunset. In that case, the calculation of u* proceeds as
for nighttime, as described in the next section.
11.2.3.2 Nighttime Heat Flux Calculation The method used under stable conditions
follows Venkatram (1980), with a first estimate of * based on an empirical equation due
to van Ulden and Holtslag (1983):

* = 0.09 1 0.5nc2

(11.29)

This states basically that * is independent of u* and H. However, they show that (11.29)
is subject to constraints approximated by L 10 m and H 60 Wm 2 . The calculation
proceeds by computing u* using the estimated * :


2 12


(11.30)




2U o
C Dn U ref

1+ 1






u* =

C Dn U ref

2
where the neutral drag coefficient, CDn is defined as ln( zref / z o ) and U 0 = z ref g * T ,
being the constant 4.7 used in equation (11.7). Then L and H are calculated, the latter
assuming acp equals 1200 J m3 K1. If L is less than 10m, L is set to 10m and u* recomputed using (11.3). If H 60 Wm 2 , i.e., u* * 0.05 Kms 1 , then u* is recalculated from a cubic equation derived by substituting this limit into equation (11.3) and
using the relation (11.13). Thus,
3
2
u* CDn Uref u* + CDn zref u* * 0 g T = 0

where u* *

(11.31)

= 0.05 Kms 1 .

It should be noted that the heat flux itself is not of primary importance here,
although it can be computed given * and u* . The friction velocity, u* , is the key
quantity since it is used to set the mean velocity profile near the surface, in scaling the
turbulence profiles and in determining the boundary layer height in the CALCULATED
mode.

144

DRAFT
11.2.4 Boundary Layer Height.

As with the surface heat flux, the planetary boundary layer height may be
specified three ways depending on the PBL mode:
1)

0BSERVATION

through the meteorology observation files

2)

as a simple diurnally varying function with a prescribed daytime

SIMPLE
maximum and nighttime minimum

3)

CALCULATED using a model that accounts for convection and shear-stress

induced mixing as well as entrainment of an overlying stable layer.

When the OBSERVATION PBL mode is used, the boundary layer height can be
specified directly in the observations input files. However, if the PGT class or L is given
without specifying a boundary layer height, one will be assumed as follows: z i = 1000 m
under neutral or unstable conditions and z i = 5 L for stable conditions. The assumed
boundary layer heights corresponding to the PGT stability classes are shown in Table
11-2.
In the SIMPLE PBL, the boundary layer height is given by

Rz
|
|S

+ z i max z i min
i min

z i = z i max

|| z
T

j a t 6f
6

,6 t 12
,12 t 18

(11.32)

, otherwise

i min

where z i min and zi max are the minimum and maximum daily boundary layer depths,
respectively. This equation approximates typical daily boundary layer evolution by
setting the depth to a minimum value during nocturnal hours, maintaining a linear growth
during morning hours, and holding a constant depth during afternoon hours. Typical
values for z i min range from 30m to 100m while zi max might range from 300m on an
overcast day to 2000m on a sunny summer day.
In CALCULATED PBL mode, SCIPUFF computes boundary layer height using
the H, u* , L and other parameters. Again, the method differs for stable and unstable
conditions.

145

DRAFT
11.2.4.1 Daytime Boundary Layer Height Calculation If the sensible heat flux is
positive, SCIPUFF solves an evolution equation for the boundary layer height. Usually,
the dominant mechanism driving the boundary layer growth is entrainment of the
overlying air by convective eddies resulting from the surface heat flux. However, during
early morning hours, surface shear stress induced (mechanical) mixing may be important.
The model includes simple parameterizations of these mechanisms but ignores
subsidence, latent heat effects, radiation and horizontal advection.
The boundary layer is assumed to be capped by a thin interfacial layer separating
it from the stable air above. The temperature within the boundary layer is assumed to be
uniform. The heat flux at the interfacial region due to entrainment, Hi, is assumed to be
proportional to the surface heat flux, following Carson (1973) and Deardorff (1980):

H i = AH

(11.33)

where A = 0.2. It can be shown (Carson, 1973) that the growth rate due to convective
mixing is

FG dz IJ
H dt K

conv

a 1 + 2 Af H

(11.34)

zi i c p

where i is the potential temperature gradient at zi .

Mechanical mixing is modeled assuming no surface heat flux and the entrainment
flux is given by (Kato and Phillips, 1969)

Hi = B

3
T u*
cp
g zi

(11.35)

where B = 2.5. Then, analogous to (11.34), we define

FG dz IJ

3
Tu*
= 2B
g i zi2

H dt K

mech

(11.36)

SCIPUFF assumes that the two mechanisms are independent and that the stronger
one dominates. Thus,

(11.37)

conv


mech

d zi
dt

d zi
dt

d zi
= max
dt

This equation is used at every large timestep, tL, to update zi.

146

DRAFT
11.2.4.2 Nighttime Boundary Layer Height Calculation Under stable conditions, zi is
found from an equation due to Nieuwstadt (1981) which interpolates between
Zilitinkevich's (1972) expression for very stable conditions and the neutral case solution
of 0.3u* f :
zi 0.3 u* f L
=
L 1 + 1.9 zi L

(11.38)

which is a quadratic equation for zi . (11.38) is appropriate for a boundary layer where the
surface stress is balanced by a constant geostrophic flow above. However, it would
typically take three to six hours to reach this equilibrium (Wyngaard, 1975), implying that
the predicted height will not often be observed since atmospheric conditions seldom
remain constant for that length of time. Therefore, under stable conditions, zi is not
allowed to exceed 5L. This limit is consistent with the LES results of Brown et al. (1994)
which show zi 5 L for a range of the stability parameter = u f L , the upper limit
corresponding to the larger values of . This limit also makes sense considering that
z L represents the height where shear-driven turbulence production is balanced by
turbulence destruction due to the buoyancy flux. Since turbulence will be increasingly
damped with height, it is appropriate to limit the boundary layer depth to be not very much
greater than L.

11.3 PBL MEAN PROFILES.

The mean velocity profile in the PBL depends on height, especially with respect
to the "surface layer height". The surface layer is the region of the boundary layer next to
the surface where the fluxes of momentum and temperature are essentially constant.
Here Monin-Obukhov similarity is valid, which implies that the velocity and temperature
profiles are functions only of the surface friction velocity u* , the surface temperature flux
H, the surface roughness z0 and the normalized height above the surface, z L . Below zs,
the surface layer height, SCIPUFF utilizes the similarity profile to determine the mean
velocity, i.e., u( z ) = u* ln z z0 m . The definition of zs is given by

b gb

z s = s zi

(11.39)

where
L0

0.1

s = 0.1 + 0.9 z i L
zi L + 5

L>0

147

(11.40)

DRAFT
The factor of 0.1 for neutral and convective conditions is a rough estimate used by many
investigators, e.g., Schumann (1988) and Wyngaard (1988). Sykes, Henn and Lewellen
(1993b) show that the surface layer depth under free convection conditions is dependent
on surface roughness and can be much less than 10% of the mixed-layer depth. However,
the presence of a mean wind is assumed here, and the resulting turbulent energy
production by the mean shear will increase the surface layer depth so that the 0.1 factor is
a reasonable estimate. The asymptotic limit for the very stable case is 1, i.e., the surface
layer extends over the entire (but typically shallow) boundary layer. Furthermore, in
keeping with the limit on zi under stable conditions (see Section 11.2.4.2), zs is never
allowed to exceed 5L.

us = uref

fSL z s ; z0 , L

fSL z ref ; z0 , L

The velocity at the top of the surface layer, us, is set to uref if zs
it is adjusted according to

zref ; otherwise

z ref < z s

(11.41)

where the surface layer profile function is defined as

fSL z ; z0 , L = ln

z + z0
m z L
z0

a f

(11.42)

and the functional dependence of m on height is emphasized. Note that if zs zref, then
zs is used in equation (11.3) instead of zref to calculate u* . Thus, zs must be found
iteratively under stable conditions since it depends, through (11.39), on L, which depends
on u* , which in turn may depend on zs through (11.3).

Figure 11-1 shows a schematic illustration of the idealized wind profiles used in
SCIPUFF depending on the relative position of zref and zs. Note that if zs zref, the
velocity profile is held constant from zref down to zs. Below zs,, the similarity profile is
used, i.e.,

u( z ) = us

b
b

fSL z ; z0 , L
,
fSL z s ; z 0 , L

z < zs

(11.43)

These particular examples show a constant velocity over the entire boundary layer
above the surface layer. This is the case only if fewer than two upper air levels are
located below zi; otherwise the velocity is interpolated from the upper air fields.

148

DRAFT

zref > zs

zs > zref

zi

zi

us

zs

uref

zref

uref

zref

us

zs

Figure 11-1. Schematic illustration of the idealized velocity profile in the planetary
boundary layer.

The mean temperature is determined by simple interpolation. Surface similarity is

used only in defining the potential temperature gradient, z , under stable
conditions, where

(11.44)

0.74
4.7
+
max( z s , zi 2) L

*
=
z

The gradient is assumed to be zero for neutral and convective conditions, since the
temperature is well-mixed except in a thin superadiabatic layer at the surface.

11.4 VEGETATIVE/URBAN CANOPY PROFILES.

As part of the surface land-cover description, the detailed profile within a canopy
layer can be described. In this context, a canopy is considered to be an extended layer
next to the surface containing a distribution of solid obstacles which the air must flow
149

DRAFT
around. This includes vegetative canopies, such as crops or forests, and also urban
canopies, with collections of buildings forming the flow obstruction.
The canopy parameters are specified on the same surface grid as terrain elevation,
and include both the canopy height and the canopy flow parameter which determine the
profile shape within the canopy. The velocity profile is assumed to be exponential, since
this shape has been determined to give a good representation for a wide range of
experimental data (Cionco, 1972). The data include both isolated obstacles, such as pegs,
in addition to vegetative canopies. Also, Macdonald et al. (1998) have shown that the
exponential profile shape is a reasonably good approximation for moderate building
densities in an urban canopy.
The canopy velocity profile is usually given as
u z = u hc fc z , for z hc

af b g af

(11.45)

where the canopy profile function is

(11.46)



z
hc

fc z = exp c 1



Here, hc is the canopy height, and c is the canopy parameter. Above the canopy, the
velocity profile is as described in Section 11.3, but is shifted upward in the vertical by the
displacement height, z d , so that
u( z ) = us

fSL z zd ; z0 , L
,
fSL z s z d ; z0 , L

b
b

hc z < z s

(11.47)

In order to provide a smooth transition across the canopy top, the canopy profile
and the surface layer profile are blended together, so that within the canopy we assume

 '#
%
 " !

%
&$"!
#

u( z ) = u(hc ) Fc fc z + 1 Fc

fSL z
fSL hc

z hc

(11.48)



where the velocity at the canopy top, hc , is given by (11.47). The "blending" function, Fc,
depends on the canopy parameter, c , and is given by

FG

Fc ( c ) = 1 exp

2
c
2 +0 c
0

IJ
K

150

(11.49)

DRAFT
where 0 = 0.5 . The function Fc varies smoothly from 1.0 to 0.0 as c increases from
zero, and allows the canopy profile to change smoothly into the undisturbed surface layer
profile as the canopy parameter tends toward zero. The displacement effect of the canopy
is also a function of c , and is simply modeled as
z d = 0.7 hc Fc ( c )

(11.50)

giving a limiting displacement of 70% canopy height for a dense canopy.

The final profile shapes are illustrated in Figure 11-2, where they are compared
with the experimental data of Cionco (1972).

2.0
wheat,
c = 2.45
corn,
1.97
xmas trees,
1.06
pegs,
0.79
bushel baskets, 0.36

1.5
z / hc
1.0

=2

=4

c=

c = 0.5

=0

0.5

0
0

0.2

0.4

0.6

0.8

u / u (2hc)
Figure 11-2. Canopy velocity profile shapes from (11.47) and (11.48) compared with the
experimental data of Cionco (1972).

151

DRAFT
SECTION 12
TURBULENCE SPECIFICATION
12.1 PLANETARY BOUNDARY LAYER.
The small-scale turbulence of the planetary boundary layer is typically described
analytically in terms of surface heat and momentum fluxes, although SCIPUFF can use
measured turbulence profiles through the observational meteorology input format
(Section 13.2.1). There are two idealized situations where turbulence profiles are well
understood, namely neutral conditions and free convection conditions.
In neutral flow, the surface heat flux is zero, or at least is negligible in comparison
with the mechanical shear effects. Free convection occurs under very light wind
conditions with a positive surface heat flux, and the turbulence is proportional to the
convective velocity scale, w* . For the idealized situations, we can represent the vertical
profiles of the turbulence velocity correlations using analytic shape functions as

2
uiu S = u* Fij z z i
j

(12.1)

for the shear-driven component, and

(12.2)

2
uiu B = w* Gij z z i
j

for the buoyancy-driven component. Note that we only use the lateral ( i = 2, j = 2 ) and
vertical ( i = 3, j = 3 ) components of the boundary layer velocity fluctuations in the closure
model described in Section 2.1.3.
Neutral profiles are relatively close to simple linear functions. Wind tunnel
studies of the aerodynamic boundary layer (Klebanoff, 1955; Townsend, 1976), and
numerical calculations of the neutral Ekman layer (Spalart, l989, Mason and Thompson,
l987) both support velocity variance profiles proportional to 1 z z i . Appropriate
dimensionless profiles are

F22 = 2.5 1 z zi
F33

a f
= 1.5 a1 z z f

(12.3)

152

DRAFT
Free convection profiles are available from the laboratory experiments of Deardorff
(1970), and Large Eddy Simulation (LES) calculations (Mason, l989; Schmidt and
Schumann, l989). A reasonably good representation is provided by the expressions
G22 = 0.13 1 + 1.5 e z z

G33 = 11 z zi
.

a f a 1.05 z z f

(12.4)

These representations approximate the boundary layer turbulence profiles under

idealized conditions. In general, the boundary layer is characterized by a spectrum of
turbulence energy over a range of eddy scales. This spectrum is modeled as made up of
two distinct turbulence populations both governed by the Kolmogorov k 5 3 law: sheardriven, near-surface turbulence with energy given by (12.3) and convection-driven
turbulence with energy given by (12.4). The latter type of turbulence is characterized by
large eddies on the scale of the mixed-layer depth, so a characteristic length scale for the
horizontal components of the buoyancy-driven turbulence is

B = 0.3 zi

(12.5)

The appropriate length scale for the shear-driven turbulence is related to distance from the
surface and is modeled very simply as
1
1
1
=
(12.6)
2 +
2
2
S
0.65 z
0.3 zi

g a

providing a transition from the linear behavior near the surface to a constant value in the
mixed layer. The vertical length scale, V , is equal to S . For stable conditions, the
convective velocity scale is zero and only the neutral profile shape (12.3) with scale (12.6)
should be used. However, the surface friction velocity will generally be reduced by the
presence of stable stratification near the surface. In practice, the dispersion model uses the
lateral component, u2 2 , for both horizontal components, since we do not account for the

cross-correlation for the boundary layer turbulence and the lateral dispersion is most
important in actual applications.
Numerous experimental measurements and LES have shown that the heat flux
profile can be represented as a linear function
w = H 0 1 z zi
B

(12.7)

153

DRAFT
The turbulence specifications given above are all based on boundary layer
parameters, and are valid only within the surface-driven boundary layer, i.e., z z i . For
situations where the plume is released above the surface boundary layer, the turbulence
conditions must be specified differently. At present, turbulence conditions in the stable
atmosphere above the boundary layer are poorly understood, so we can only suggest a
very crude representation. A typical value for the vertical velocity variance, w B2 , is

2s2 and a typical vertical length scale, , is 10m. We suggest using these values
0.01m
V
for all locations above the boundary layer. The horizontal fluctuations are more difficult
to specify, since horizontal motions in a stably-stratified fluid can persist long after their
generation by flow over terrain or wave-breaking events. Hanna (1983) and Etling
(1990) discuss some of the possible mechanisms for horizontal meandering under stable
conditions and show that there is typically a r.m.s. lateral velocity fluctuation between 0.5
and 1ms1, even under very light wind conditions. The time scales of these fluctuations
indicate a horizontal length scale of the order of 1000m. We therefore suggest that
horizontal velocity variances, u B2 and v B2 , be taken as 0.25m2s2 and the horizontal

scale, B, as 1000m outside the boundary layer.

In applying the analytical turbulence profiles, the stable values for the horizontal
and vertical velocity variances are used as lower limits for the boundary layer values. If
the surface generation mechanisms imply a smaller horizontal variance, then a secondary
component is added to the B-component with the stable length scale and the composite
horizontal length scale is calculated as the energy-weighted average of the two scales.

12.2 VEGETATIVE/URBAN CANOPY LAYER.

In conjunction with the velocity profile assumptions for the vegetative/urban
canopy layer in Section 11.4, we also specify modified profiles for turbulence quantities.
Experimental observations of turbulence profiles within either vegetative or urban
canopies are very limited, and display significant scatter, e.g., Raupach et al. (1996). For
the present, we assume a very simple linear profile for the r.m.s. velocity fluctuations,
reducing from the external value at the canopy top to a density-dependent near-surface
value. Thus,
u 2 z = fct ( z ) uB2 z , z < hc

Bc

where, uB2 is defined from (12.2), and the canopy turbulence profile function is

154

(12.8)

DRAFT

FG

fct ( z ) = 1 1

IJ c

z
1 e 1.5
hc

(12.9)

The same reduction factor is applied to all boundary-layer scale turbulence

components within the canopy layer. Figure 12-1 illustrates the profile shapes for a range
of canopy parameter values corresponding to the velocity profiles shown in Figure 11-2.
The boundary layer depth is assumed to be large in comparison with the canopy height
(zi hc = 100), so that the non-canopy ( c = 0) profiles are effectively constant over the
range of heights in the figure.

12.3 MESOSCALE/SYNOPTIC SCALE.

Large-scale variability is assumed to be constrained to the horizontal plane with no
vertical heat flux or length scale. Therefore, the only parameters needed for large-scale
characterization are the velocity fluctuation correlations u L2 , v L2 , and u v L and the

horizontal length scale HL . These parameters can be specified by model, as

meteorological input, or as direct input.
The large-scale variability model used in SCIPUFF is based on research by
Nastrom and Gage (1985) and a theoretical model proposed by Gifford (1988). Nastrom
and Gage analyzed GASP (Global Atmospheric Sampling Program) wind data and
showed a clear spectral break at a horizontal wavelength of about 400 km, with k3
spectrum at longer wavelengths and k5/3 for shorter scales. Gifford argues that the largescale behavior is consistent with the predictions of two-dimension turbulence theory,
while the k5/3 behavior indicates three-dimensional motions in that part of the spectrum.
We assume that the upper-troposphere data is representative of the entire lower
atmosphere, although there is probably some vertical variation in reality. However, no
data presently exist to define these fluctuation profiles, so, the simplest assumption is
made. Furthermore, the GASP data represent long-term averages and cannot provide any
relationship between the local meteorological conditions and the local smaller-scale
velocity fluctuations. Use of the GASP data is therefore equivalent to a climatologically
averaged spectrum. Using Gifford's suggestions and the GASP profiles, which show an
increase in fluctuation variance toward the poles, the horizontal velocity variance is
represented as

155

DRAFT
2.0
(a)
1.5

z / hc
1.0

0.5

=4

=2

0.4

=1

v / u
*

=0

= 0.5

0.8

1.2

1.6

2.0
(b)
1.5

z / hc
1.0

=2

0.5

=4

0.4

=1

w / u
*

0.8

=0

= 0.5

1.2

Figure 12-1. Turbulence profiles within the canopy layer for different values of the
canopy parameter, c . (a) horizontal turbulence (shear driven component),
(b) vertical turbulence component.

156

DRAFT
cos 75 ) ,

(f

< 75
75

( f0 cos ) ,

uT2 = vT2 =

(12.10)

where is the latitude, f 0 is the twice the rotation rate of the earth, and T is the average
tropospheric energy dissipation rate. A value of T = 4104 m2s3 is suggested by
Gifford, but the inverse cosine variation is an empirical factor to provide a match with the
latitudinal variation observed in the GASP data. The turbulence length scale associated
with these dissipative fluctuations is taken to be
12

T = uT 2 + vT 2

f 1

(12.11)

where f is the Coriolis parameter, defined as

sin ,
sin15 ,

> 15
15


(12.12)

f = f0

The limit of 15 is necessary to avoid the singularity in the definition of T.

The velocity variance used in the dispersion calculation depends on the resolution
of the input wind fields since the turbulence input represents only the unresolved
component of the wind. We therefore define a filter scale, G , obtained from the wind2
2
field grid spacing (xG, yG) as G = 1.4 xG + yG , then
HL = min G , T

(12.13)

and
HL
T

23




(12.14)


u L = uT

consistent with a k5/3 spectral behavior. The cross-correlation, u v L , is assumed to be

zero in the model.

The large-scale variability parameters, u L2 , v L2 , u v L and HL , can also be

specified explicitly or read from the SURFACE and/or PROFILE observation files if
available. Only the velocity variances can be read from the observation file; the
turbulence length scale must be given explicitly. Velocity variances read in from
observation files are interpolated in the same fashion as the mean fields; see Section 10.4.

157

DRAFT
Near the surface, the large-scale velocity variances behave similarly to the mean
wind, and are therefore scaled using the velocity profile shape functions. The variances
are reduced from the free atmosphere values within the surface layer, using the surface
layer shape function, fSL . Within a surface canopy layer, the additional reduction due to
the canopy profile function is also incorporated.

12.4 METEOROLOGICAL UNCERTAINTIES.

Characterizing the state of the atmosphere always entails a degree of uncertainty.
Even assuming that perfect measurements or gridded analysis fields are available, there
is uncertainty in interpolating to other locations and times. Of course, numerical weather
predictions grow more uncertain with forecast time, even when starting from accurate
initial conditions. Characterizing the meteorological uncertainty depends on a number of
factors including the data quality, spatial and temporal separation from observations or
calculation points as well as features of a particular locale such as terrain or proximity to
water bodies that may induce quasi-deterministic flows. However, a full treatment of
these factors is beyond our current capabilities. Here we attempt only to make plausible
quantitative estimates of the uncertainties inherent in meteorological data that can be
utilized within the probabilistic framework of SCIPUFF, as described in Section 9.4.
The general idea is to treat uncertainty in a fashion similar to large-scale
variability. The further away in time or space from an observation or grid location, the
greater the inherent uncertainty or variance, with the asymptotic limit being the
climatological variance. If suitable statistics characterizing the uncertainty of the
meteorological input are available, these may be used directly through the observation
file format described in Section 13.2.1. Some examples might be error estimates
generated by numerical models as a function of space and time or long-term wind
statistics given in terms of means and variances. However, this information is rarely
available and therefore we must construct appropriate models of meteorological
uncertainty.
Studies on error growth in numerical weather prediction models, e.g., Leith
(1978), have generally found that the r.m.s. error-doubling time is about 2 to 3 days,
assuming small initial errors, with a shorter time appropriate for a null model or
persistence assumption, i.e., no changes from the initial state. So, in the case of forecast
158

DRAFT
models, we make a crude estimation of error growth using a function that interpolates
between fairly rapid error growth on the timescale of a few days and slow asymptotic
growth to the climatological variance:

Eu

Eu (t ) = Eu0 + f E t,

fc

(12.15)

where Eu is the error variance for the east-west wind component, Eu0 is the initial error,
Eu is the climatological variance and t fc is a timescale appropriate to error growth in
numerical forecast models which we set to 3 days. Here, t is time from an initial analysis
time (which is assumed to be the first time in a gridded meteorology input file or specified
in the header information of a special PROFILE file, Section 13.2.1). The S-shaped
interpolating function f E is given by

F t IJ
f at, f = 1 expG
H a + t f K
2

(12.16)

Error growth under the assumption of persistence is modeled with the same functional
form as (12.15), but with a timescale of t ps = 1.5 days. The normalized error growth
curves for forecast and persistence models, i.e., function f E evaluated with t fc and t ps , are
shown in Figure 12-2.

Separate calculations analogous to (12.15) of the north-south velocity variance Ev

and uv-correlation Euv are made if the corresponding climatological values Ev and Euv

are available. SCIPUFF utilizes a climatology data base containing Eu , Ev and Euv as
functions of latitude, longitude, altitude, month and time of day. Linear interpolation to
the meteorology input grid and time is then employed. If the data base is not available,

we set Ev = Eu and Euv = 0 and assume a linear variation in Eu from the surface to the
top of the troposphere:

,0 ETrop E BL

z trop

z trop z

Eu = E BL + min 1

(12.17)

where E BL and ETrop are values typical of the atmospheric boundary layer and tropopause,
respectively, and z Trop = 10000 m represents the top of the troposphere. We tentatively set

E BL = 5 ms 1 and ETrop = 12 ms 1 based on an analysis of wind fields from the National

Meteorological Centers Nested Grid Model for North America (Sykes, et al., 1993c).

159

DRAFT
1.0

ps = 1.5 days

0.8

Eu (t ) E
Eu

fc = 3 days

0.6

0
u

0.4

0.2

10

t (days)

Figure 12-2. Normalized error variance growth for forecast and persistence assumptions.

A length scale E is also associated with the meteorology uncertainty. Again the
functional form of (12.15) is used so that E is found from an interpolation between the
climatological value and the initial length scale 0E = min G , . The
E
E
climatology length scale is determined as a function of the total variance and altitude.
Based on the discussion in the Section 12.3, we use a k5/3 assumption to extend the
Gifford spectrum and set a climatological length scale representative of the troposphere
as

= T 2
0

MM

+ Ev

32

h OP
PP

uT 2

LM 1 c E

(12.18)

However, we expect shorter length scales near the surface, especially below the mixed
layer height. Therefore, we set by interpolating between and a length scale
0
E
BL
representative of the atmospheric boundary layer:
= + 1
0
E
BL

(12.19)

where
1
1 + max z z BL ,0 z BL

160

(12.20)

DRAFT
We tentatively set = 100 km based on a case study using observational data from a
BL
365 day period over the Korean peninsula. We choose z BL = 1000 m as a representative
mixed-layer height and a transition depth of z BL = 200 m .

The initial variance is found by scaling the climatological variance down to the
initial length scale,

0E
E =

0
u

23

Eu

(12.21)

0
with Ev0 and Euv determined similarly.

A slight modification to (12.15) is required when t is beyond the last available

forecast time since the variance grows up to that time on the timescale fc , while,
subsequently, it grows on the timescale ps . In this case we have
E u (t ) = Eu (t fc ) + f E t t fc , ps Eu

(12.22)

where t fc is the last forecast time (relative to the analysis time) and Eu (t fc ) is found from

(12.15). Similar modifications are made for

and Ev , and Euv , if appropriate.

Note that equations (12.15) or (12.22) are applied at grid locations. When
interpolating to puff centroids, an additional variance should be included to account for
the uncertainty due to subgrid motions, but this effect is probably small and currently
ignored.
Modeling the uncertainty associated with observational meteorology is
conceptually similar but we also include spatial separation from the nearest observation
location. Here, we work on a three-dimensional grid that is generally of finer resolution
than the one described in Section 10.1 for the mean field interpolation. The observation
nearest to any location on this grid is taken to be the one that produces the greatest weight
as defined in (10.2). The weight is then converted into a horizontal distance based on
(10.3), so that
1
2
rmin = a 2 w max 1


(12.23)

where the core radius a is defined as in Section 10.2 using the finer grid spacing. wmax is
the maximum weight from the SURFACE or PROFILE observations (but is never greater
than unity) and is a function of horizontal separation, vertical separation and possibly

161

DRAFT
temporal separation as given by (10.6). We then use rmin and in f E to estimate the
E
uncertainty variance on the grid as
~
Eu0 = f E rmin , Eu
E

(12.24)

~
~0 ~
and similarly for Ev0 and Euv . Eu0 can be evaluated at each observation time t obs , but since
we generally are interpolating wind fields between observation times, we must also
account for added uncertainty due to temporal separation. The total variance associated
with a given observation time is therefore
~
~

Eu t = Eu0 + f E t t obs , ps Eu
(12.25)

where the persistence timescale is used. The total variance at time t, Eu (t ), is then found
~
by linear interpolation between the values of Eu (t ) evaluated at the two bracketing
observation times. Obviously, no interpolation is required if t is before the first
observation time or after the last one.

The length scale associated with observational uncertainty is found from the
Gifford spectrum as

LM 1 b E

E = T 2

MM
N

+ Ev

uT 2

32

g OP

(12.26)

PP
Q

Since the calculations for observational meteorology are performed on a threedimensional grid, the variance and length scale at a puff centroid are found from linear
interpolation. This means that if no grid point corresponds exactly with the observation
location, there will be a non-zero uncertainty associated with a puff even if it coincides
exactly in time and space with an observation. This dependence on the grid is certainly
not ideal, but the methodology allows an initial examination into the effects of
uncertainty. For calculations involving extrapolations of more than a few hours from
observation times, this effect is probably negligible.
As mentioned in Section 13.2.1, observational input can be characterized as
standard OBSERVATION type (just described), ANALYSIS type or FORECAST
type. The latter two are intended for use with profiles extracted from numerical models.
ANALYSIS profiles are assumed to represent the numerical model after suitable
assimilation and analysis of observations, i.e., appropriate for initializing a numerical

162

DRAFT
prediction model. In this case, the meteorological uncertainty is treated as described
above for standard observational input, with the observation locations corresponding to
the selected model grid points. Thus, the meteorology at grid locations is assumed to be
perfect. The initial uncertainty for FORECAST input is also set as for standard
observational input. This is assumed to correspond to an analysis time given in the
PROFILE file (see Section 13.2.1) or the first time on the profile file if no analysis time
is explicitly given. The subsequent time evolution is then described by equation (12.15).

163

DRAFT
SECTION 13
FILE FORMATS
13.1 INPUT FILES.
13.1.1 Input File Structure.

The input data for PC-SCIPUFF is constructed by the Graphical User Interface,
but consists of Fortran NAMELIST files. There are three input files:
ProjectName.INP

- the main input parameter file

ProjectName.SCN - the release scenario file

ProjectName.MSC - the meteorological scenario file
where ProjectName is the identifying name for the PC-SCIPUFF calculation. The files
are maintained separately so that an arbitrary number of releases can be made, and
multiple release material types can be defined.
PC-SCIPUFF input parameters are specified through namelist groups and
unformatted data located in the input, release scenario, and meteorology scenario files.
All namelist parameters must be contained within the '&groupname' and '&end' lines
of the given input file. The '&' on these lines must be preceded by at least one space in
order for the program to recognize this block as namelist input. All lines contained
outside of the namelist section are considered comments and are disregarded by PCSCIPUFF. Some input parameters have default values; if a parameter does not have a
default, the program will abort if the parameter is not included in the namelist.

13.1.2 Main Input File.

A sample input file is shown in Figure 13-1. The following section describes the
namelist parameters and their defaults, if available.

164

DRAFT
&CTRL
RESTART
&END
&TIME1
YEAR_START
MONTH_START
DAY_START
TSTART
TZONE
LOCAL
&END
&TIME2
YEAR_END
MONTH_END
DAY_END
TEND
TEND_HR
DELT
DT_SAVE
&END
&FLAGS
TITLE
CREATE
AUDIT_CLASS
AUDIT_ANALYST
&END
&DOMAIN
CMAP
XMIN
XMAX
YMIN
YMAX
ZMAX
VRES
HRES
&END
&OPTIONS
T_AVG
CMIN
LSPLITZ
DELMIN
&END
&MATDEF
MNAME
CLASS
NSG
DENSITY
PSIZE
PBOUNDS
GROUP_DEPOSITION
TOTAL_DEPOSITION
GROUP_DOSE
TOTAL_DOSE
UNITS
&END

= F,
=
=
=
=
=
=

1979,
4,
2,
5.0,
5.0,
F,

=
=
=
=
=
=
=

1979,
4,
2,
19.0,
14.0,
300.0,
300.0,

=
=
=
=

'SCIPUFF RUN',
F,
'UNCLASSIFIED,
'Anonymous',

=
=
=
=
=
=
=
=

'CARTESIAN',
-1.0,
9.0,
-9.0,
1.0,
2500.0,
250.0,
1.0,

=
=
=
=

0.0,
0.0,
T,
0.0,

=
=
=
=
=
=
=
=
=
=
=

'ANTH',
'PART',
1,
500.0,
2.00E-06,
1.99E-06 2.01E-06,
F,
T,
F,
T,
'kg',

Figure 13-1. Sample PC-SCIPUFF input file.

165

DRAFT
The namelist group input parameters are defined as follows.
Namelist: CTRL - run control flag
restart

- Flag for run status (LOGICAL). If 'True', the old project file is
used to continue the run from the last time break. The default is
'False'.

file_rst

- name of project used to initialize current project.

(CHARACTER*128). If not blank, the current project is run on
from the state defined by the file_rst project, puff and surface files.
The default is ' ' (blank).

path_rst

- path for project file_rst. (CHARACTER*128). The default is ' '.

time_rst

- time on the puff and surface files of project file_rst used to

initialize the current project (REAL*4). The default is the last
time.

Namelist: TIME1 - time domain information

year_start

- Start year of the calculation (INTEGER*4). If the year is not

specified, it is assumed to be that of the first year given in the
meteorological input.

month_start

- Start month of the calculation (INTEGER*4). If the month is not

specified, it is assumed to be that of the first month given in the
meteorological input.

day_start

- Start day of the calculation (INTEGER*4). If the day is not

specified, it is assumed to be that of the first day given in the
meteorological input.

tstart

- Start time (hours) of the calculation (REAL*4). If the time is not

specified, it is assumed to be that of the first time given in the
meteorological input.

tzone

- Local time of midnight UTC (REAL*4). The default is zero.

local

- Flag for calculation time convention (LOGICAL). If 'True', the

time is taken as local instead of UTC. The default is 'False'.

166

DRAFT
Namelist: TIME2 - supplementary time domain input used for restart
year_end

- Stop year of the calculation (INTEGER*4). If the year is not

specified, it is assumed to be the same as the start year.

month_end

- Stop month of the calculation (INTEGER*4). If the month is not

specified, it is assumed to be the same as the start month.

day_end

- Stop day of the calculation (INTEGER*4). If the day is not

specified, it is assumed to be the same as the start day.

tend

- Stop time (hours) of the calculation (REAL*4). If the time is not

specified, it is assumed to be 1 hour past the start time.

tend_hr

- Duration (hours) of the calculation (REAL*4).

delt

- Maximum timestep (seconds) of the calculation (REAL*4). The

default is 900.

dt_save

- Amount of time (seconds) between output dumps (REAL*4). The

output includes the instantaneous fields and the surface
accumulation fields, as specified by the user. The default is 6.

Namelist: FLAGS - run control and audit information

title

- Descriptive title of the project (CHARACTER*80). No default is

available.

create

- Flag for full PC-SCIPUFF calculation (LOGICAL). If 'True', PCSCIPUFF only initializes the project, but does not start the
calculation. The default is 'False'.

audit_class

- Classification of the current project (CHARACTER*32). No

default is available.

audit_analyst

- Name of the project analyst (CHARACTER*32). The default is

'anonymous'.

dynamic

- Flag to including momentum and buoyancy dynamics

(LOGICAL). If 'False' then puffs are treated as passive. Only
gaseous puffs can carry dynamics; however all puffs will be
influenced by the dynamics.

167

DRAFT
dense_gas

- Flag to include dense gas effects (LOGICAL). If 'False' then the

gaseous density parameter is ignored. If 'True', dynamics must
also be 'True'.

static

- Flag to perform quasi-steady calculation of continuous sources

(LOGICAL). The default is 'True'.

run_mode

- Flag to control run parameters (INTEGER*4). The default is

STANDARD_MODE (value is 0). If FAST_MODE (value is 1) is
set, then a coarser resolution calculation is performed; surface grid
and vertical resolution is restricted.

multicomp

- Reserved for future implementation of multiple chemical species

(LOGICAL).

hazarea

- Type of hazard area prediction.(CHARACTER*4). The three

recognized types are 'OFF' for no calculation, 'ON' for a dispersion
calculation including meteorological uncertainty and 'COMB'
which calculates dispersion with and without meteorological
uncertainty simultaneously. The default is 'OFF'.

Namelist: DOMAIN- defines the spatial domain

cmap

- Type of spatial domain, 'UTM', 'CARTESIAN' or 'LATLON'

(CHARACTER*40). The default is 'LATLON'.

xmin

- Minimum horizontal coordinate (REAL*4). For a LATLON run,

this value represents the minimum longitude of the domain in
degrees. For CARTESIAN or UTM runs, this value represents the
western boundary in kilometers.

xmax

- Maximum longitude (deg) or eastern (km) boundary (REAL*4).

ymin

- Minimum latitude (deg) or southern (km) boundary (REAL*4).

ymax

- Maximum latitude (deg) or northern (km) boundary (REAL*4).

zmax

- Vertical extent (meters) of the calculation domain (REAL*4). The

default is 2500.

vres

- Spacing parameter (meters) that limits the vertical growth of a puff

(REAL*4). The default is 250.

168

DRAFT
hres

- Spacing parameter that limits the horizontal growth of a puff in the

same units as xmin (REAL*4). For observational data, the default
is 1/10 of the domain. For gridded data, the resolution is taken as
that of the meteorology.

UTM_zone

- UTM reference zone when UTM coordinates are specified

(INTEGER*4).

xref

- Cartesian x-coordinate (km) of the reference point when

CARTESIAN or UTM coordinates are specified (REAL*4). The
default is zero.

yref

- Cartesian y-coordinate (km) of the reference point when

CARTESIAN or UTM coordinates are specified (REAL*4). The
default is zero.

lon0

- Longitude of the reference point in degrees, when CARTESIAN or

UTM coordinates are specified (REAL*4). The default is zero and
east is positive.

lat0

- Latitude of the reference point in degrees, when CARTESIAN or

UTM coordinates are specified (REAL*4). The default is zero and
north is positive.

domain_status

- Reserved for Incident Source Models (INTEGER*4 array).

Namelist: OPTIONS - optional parameters

t_avg

- Conditional averaging time (seconds) for defining the diffusive

component of turbulence (REAL*4). The default is no conditional
averaging.

cmin

- Minimum puff mass (REAL*4). The default is zero.

lsplitz

- Flag for vertical puff splitting within the planetary boundary layer
(LOGICAL). If 'True', puffs are not split in the vertical direction
within the boundary layer. The default is 'False'.

delmin

- Minimum grid size (meters) for the adaptive surface grid

(REAL*4). The default is zero.

169

DRAFT
wwtrop

- Tropospheric vertical turbulent fluctuations (m2s2) used as the

minimum value for vertical velocity variance (REAL*4). The
default is 0.01.

sltrop

- Tropospheric vertical length scale (meters) (REAL*4). The

default is 10.

epstrop

- Tropospheric energy dissipation rate (m2s3) (REAL*4). The

default is 4.0104.

uu_calm

- Minimum horizontal velocity fluctuation variance (m2s2)

(REAL*4). The default is 0.25.

sl_calm

- Horizontal length scale associated with uu_calm (meters)

(REAL*4). The default is 1000.

nzbl

- Number of boundary layer vertical grid levels (INTEGER*4). The

default is 11.

mgrd

- Grid resolution parameter that limits the horizontal growth of a

puff (INTEGER*4). The default is 2.

grdmin

- Minimum grid size (meters) for the puff grid resolution (REAL*4).
The default is zero.

z_dosage

- Elevation at which surface dosages are computed (REAL*4).

Default is 0.0

smpfile

- Name of a file with sampler locations for sampler output

(CHARACTER*80). See Section 13.1.5.

dt_smp

- Minimum output time (s) to sampler file (REAL*4). Default is

large time step, delt.

Namelist: MATDEF - defines material properties (multiple copies of this

namelist can be included in the input file)
mname

- Material identifying name (CHARACTER*16).

class

- Material class (CHARACTER*4). Must be 'GAS' (gaseous),

'PART' (particle), 'LIQ' (liquid), or AERO (flashing
liquid/aerosol) for a user-defined material. Special material class
names are reserved for Incident Source Models (see Section 8);
these are NWPN for a nuclear weapon material, and NFACG,
170

DRAFT
NFACP, NFACL, NFACA for nuclear facility materials of type
gas, particle, liquid, flashing aerosol, respectively. The generic
nuclear facility class, NFAC, represents a gaseous material.
nsg

- Number of particle or liquid droplet size groups (INTEGER*4).

Only used for particle and liquid type materials.

density

- Density of gaseous or particle material. Vapor phase density of

liquid or flashing aerosol material (kg/m3) (REAL*4).

liquid_density

- Two liquid material liquid phase density parameters a and b

(REAL*4). The liquid phase density (kg/m3) is defined as abT.

antoine

- Three liquid material vapor pressure coefficients a, b and c

(REAL*4). The vapor pressure is defined as
log 10 ( Pv ) = a b / (T + c ) where Pv is in mm Hg and T in C.

mweight -

- Liquid or flashing aerosol material molecular weight (REAL*4).

surf_tension

- Liquid material surface tension (N/m) (REAL*4)

spread_factor

- Liquid material secondary evaporation parameter (REAL*4). It is

defined as the ratio of the droplet diameter as it sits on the surface
to the diameter before impact. If this parameter is less than or
equal to zero then no secondary evaporation is computed.

gas_deposition

- Gas or vapor deposition velocity (m/s) (REAL*4). Not used for

particle materials.

psize

- Array of particle or liquid droplet mass mean diameters (meters)

(REAL*4). Only used for particle and liquid type materials.

pbounds

- Bin boundaries (meters) that determine the range of particle or

liquid droplet sizes within the subgroup (REAL*4). Only used for
particle and liquid type materials.

group_deposition - Flag for group deposition (LOGICAL). If 'True', the deposition of

each subgroup is included on the output surface deposition file.
The default is 'False'.
total_deposition - Flag for total deposition (LOGICAL). If 'True', the total
deposition of all subgroups is included on the output surface
deposition file. The default is 'False'.

171

DRAFT
group_dose

- Flag for group dose (LOGICAL). If 'True', the dose of each

subgroup is included on the output surface dose file. The default is
'False'.

total_dose

- Flag for total dose (LOGICAL). If 'True', the total dose of all
subgroups is included on the output surface dose file. The default
is 'False'.

units

- Material mass units for labeling plots (CHARACTER*16).

conc_min

- Minimum concentration of interest for this material (REAL*4).

evap_min

- Minimum concentration for creating puffs due to surface

evaporation (REAL*4). Only used for liquid type materials.

NWPN_decay

- Decay exponent of nuclear weapon (NWPN) material activity

(REAL*4). The default is 1.3.

decay_amp

- Daytime decay rate (s1) of the material activity (REAL*4). The

default is zero.

decay_min

- Nighttime decay rate (s1) of the material activity (REAL*4). The

default is zero.

specific_heat_liq - Liquid phase specific heat (J kg1 K1) for flashing aerosol
material (REAL*4). The default is 1000.
specific_heat_vap - Vapor phase specific heat (J kg1 K1) for flashing aerosol material
(REAL*4). The default is 1000.
multi_comp

- Reserved for future implementation of multiple chemical species

(LOGICAL).

file_name

- Name of file containing default contours for output display

(CHARACTER*128). Not used in dispersion calculation.

file_path

- Path of file containing default contours for output display

(CHARACTER*128). Not used in dispersion calculation.

13.1.3 Release Scenario File.

A sample release scenario file is shown in Figure 13-2. The following section
describes the namelist parameters and their defaults, if available.

172

DRAFT

&SCN
TREL
XREL
YREL
ZREL
RELTYP
CMASS
SIZE
RELMAT

=
=
=
=
=
=
=
=

0.0,
0.0,
0.0,
250.0,
'I',
10.0,
10.0,
'DUST',

&END

Figure 13-2. Sample PC-SCIPUFF release scenario file.

The namelist parameters are defined as follow.

trel

- Release time (hours) (REAL*4). No default is available.

xrel

- Absolute x-coordinate of the release, Cartesian (km) or Lat/Long

(degrees) (REAL*4). No default is available.

yrel

- Absolute y-coordinate of the release, Cartesian (km) or Lat/Long

(degrees) (REAL*4). No default is available.

zrel

- Release height (meters) above the surface (REAL*4). No default

is available.

reltyp

- Type of release, 'C' for continuous, 'I' for instantaneous , 'CM' for
moving or 'CP' for evaporating pool (CHARACTER*4). No
default is available.

cmass

- Total mass of material for an instantaneous release and liquid pool

release or mass flux for a continuous or moving release (REAL*4).
No default is available.

size

- Source radius or initial Gaussian spread (meters) (REAL*4). No

default is available.

tdur

- Duration (hours) of release for a continuous release (REAL*4).

No default is available.

name_rel

- Name of the CLOUDTRANS file containing the puff data for an

instantaneous release (CHARACTER*80). No default is available.

173

DRAFT
relmat

- Name identifying the material being released (CHARACTER*16).

Must be identical to a name given in the materials list in the input
file. No default is available.

sigx

- x-direction source spread (meters) (REAL*4). No default is

available. For the special case of a flashing aerosol material, sigx
contains the (dimensionless) x-direction entrainment ration, Ex (see
Section 7.4).

sigy

- y-direction source spread (meters) (REAL*4). No default is

available. For the special case of a flashing aerosol material, sigy
contains the (dimensionless) y-direction entrainment ration, Ey (see
Section 7.4).

sigz

- z-direction source spread (meters) (REAL*4). No default is

available. For the special case of a flashing aerosol material, sigz
contains the (dimensionless) z-direction entrainment ration, Ez (see
Section 7.4).

subgroup

- Number of the individual material subgroup for a particle or liquid

release (INTEGER*4). No default is available. For particle and
liquid material releases if subgroup is equal to nsg+1 where nsg is
the number of size bins defined in the MATDEF namelist in
project.inp, then the cmass is distributed among all material bins
with a lognormal distribution. For liquid materials if subgroup is 0
then the cmass is released as the vapor phase.

urel

- x-direction velocity (m/s) for a moving continuous source

(REAL*4). No default is available.

vrel

- y-direction velocity (m/s) for a moving continuous source

(REAL*4). No default is available.

wrel

- z-direction velocity (m/s) for a moving continuous source

(REAL*4). No default is available.

wmom

- Integrated vertical momentum release rate (m 4 / s 2 ) for continuous

releases and total vertical momentum (m 4 / s 2 ) for instantaneous
releases (REAL*4).

buoy

- Integrated buoyancy release rate (Cm 3 / s ) for continuous releases

and total buoyancy ( Cm 3 ) for instantaneous releases (REAL*4).
174

DRAFT
lognorm_mmd

- The mass-mean diameter (meters) of the size distribution for

lognormal releases (REAL*4).

lognorm_sigma

- The sigma of the distribution for lognormal releases (REAL*4).

number_random - The number of copies of this release to be randomly released

(INTEGER*4). Default is 1.
random_spread

- The horizontal spread range over which the releases are randomly
released (meters) (REAL*4). Default is 0.0.

random_seed

- Random generator seed value (INTEGER*4).

horiz_uncertainty - Horizontal location uncertainty (meters) assumed to be the spread

of a circular Gaussian (REAL*4).
vert_uncertainty - Vertical location uncertainty (meters) assumed to be the spread of a
Gaussian (REAL*4).
rel_mc

- Reserved for future implementation of multiple chemical species

(REAL*4).

opid

- Reserved for Incident Source Models (INTEGER*4).

opmod

- Reserved for Incident Source Models (INTEGER*4 array).

opready

- Reserved for Incident Source Models (LOGICAL).

13.1.4 Meteorology Scenario File.

A sample meteorology scenario file is shown in Figure 13-3. The following

section describes the namelist parameters and their defaults, if available.
The namelist parameters are defined as follow.
met_type

- Type of meteorological input, 'OBS', 'GRIDDED', 'FIXED',

'CLIMO', or 'CLI3D' (CHARACTER*80). No default is available.

bl_type

- Type of boundary layer input, 'SBL', 'CALC', 'OBS', 'OPER',

PROF or 'NONE' (CHARACTER*80). No default is available.

ensm_type

- Type of large-scale meteorological input, 'INPUT', 'MODEL',

'OBS', 'NONE' (CHARACTER*80). No default is available.

175

DRAFT
&MET
MET_TYPE
BL_TYPE
ENSM_TYPE
UU_ENSM
SL_ENSM
H_CNP
ZRUF
ALBEDO
BOWEN
CLOUD_COVER
LOCAL_MET
&END
1 $apr_sfc.obs

=
=
=
=
=
=
=
=
=
=
=

'OBS',
'CALC',
'NONE',
0.10,
0.10,
-1.0,
0.10,
0.14,
0.30,
1.00,
F,

1 $apr_ua.obs

Figure 13-3. Sample PC-SCIPUFF meteorology scenario file.

uu_ensm

- Velocity variance (m2s2) for calculating the large-scale

component of the dispersion (REAL*4). Used only with 'INPUT'
large-scale meteorology. The default is 0.25.

sl_ensm

- Length scale (meters) for calculating the meandering component of

the dispersion (REAL*4). Used only with 'INPUT' or 'OBS' largescale meteorology. The default is 1.0105.

sl_haz

- Length scale (meters) for calculating dispersion due to wind

uncertainty given as velocity variance on observational
meteorology input files (REAL*4). Used only when hazarea is
'ON' or 'COMB'. The default is 1.0105.

zimin

- Minimum daily inversion height (meters) for calculating boundary

layer parameters (REAL*4). Used only with 'SBL' boundary
layers. No default is available.

zimax

- Maximum daily inversion height (meters) for calculating boundary

layer parameters (REAL*4). Used only with 'SBL' boundary
layers. No default is available.

hconst

- Minimum daily surface heat flux (Wm-2) for calculating boundary

layer parameters (REAL*4). Used only with 'SBL' boundary
layers. No default is available.

hdiur

- Maximum daily surface heat flux (Wm-2) for calculating boundary

layer parameters (REAL*4). Used only with 'SBL' boundary
layers. No default is available.
176

DRAFT
h_cnp

- Canopy height (meters) (REAL*4). A negative value indicates no

vegetative canopy. No default is available.

zruf

- Surface roughness (meters) (REAL*4). No default is available.

albedo

- Fraction of incident light that is reflected by the surface

(REAL*4). Used only with 'CALC' boundary layers. No default is
available.

bowen

- Ratio of surface sensible heat flux to latent heat flux (REAL*4).

Used only with 'calculated' 'CALC' boundary layers. No default is
available.

cloud_cover

- Fractional cloud cover where complete overcast is 1 and clear sky

is 0 (REAL*4). Used only with 'CALC' boundary layers. No
default is available.

local_met

- Flag specifying the reference time for the Met files (LOGICAL).
If 'True' then it is assumed that the times in the Met files are local
times. If 'False' the times are assumed to be UTC (GMT) times.

tbin_met

- Observation time bin width in seconds (REAL*4). Observations

within a bin are all given the same time. The default is no time
binning.

pr_type

- Global precipitation type, NONE, 'LGTRAIN', 'MODRAIN',

'HVYRAIN', 'LGTSNOW', 'MODSNOW', 'HVYSNOW'. If
'METFILE' is specified, precipitation data will be read from the
meteorological input file (CHARACTER*80). Default is NONE.

nearest_sfc

- Number of nearest surface observation stations used for

meteorology interpolation (INTEGER*4). The default is all.

nearest_prf

- Number of nearest upper air profile observation stations used for

meteorology interpolation (INTEGER*4). The default is all.

lmc_ua

- Flag to calculate a mass-consistent adjustment to the wind fields

(LOGICAL). This is relevant only if met_type is 'OBS', 'CLIMO'
or 'CLI3D'. The default is 'False'.

lout_mc

- Flag to output the adjusted wind fields in a MEDOC-type file

(LOGICAL). This is relevant only if lmc_ua is 'True'. The default
is 'False'.

177

DRAFT
lformat

- Flag to output the adjusted wind fields in ASCII format

(LOGICAL). If 'False', a binary file is created. This is relevant
only if lmc_ua and lout_mc are 'True'. The default is 'False'.

file_ter

- Name of file containing terrain information (CHARACTER*80).

This is required if lmc_ua is 'True'. No default is available.

alpha_max

- Maximum value of vertical adjustment parameter for massconsistent wind field calculation (REAL*4). This is relevant only
if lmc_ua is 'True' and MC-SCIPUFF is invoked. (It has no effect
on a SWIFT calculation.) The default is 1.

alpha_min

- Minimum value of vertical adjustment parameter for massconsistent wind field calculation (REAL*4). This is relevant only
if lmc_ua is 'True' and MC-SCIPUFF is invoked. The default is 0.

max_iter_ac

- Maximum number of iterations allowed for MC-SCIPUFF

calculation using the point relaxation method (INTEGER*4). The
default is 10000.

max_iter

- Maximum number of iterations allowed for MC-SCIPUFF

calculation using the FFT method (INTEGER*4). The default is
30.

ac_eps

- Convergence criterion for MC-SCIPUFF calculation using the

point relaxation method (REAL*4). The default is 1.010 2.

p_eps

- Convergence criterion for a MC-SCIPUFF calculation using the

FFT method (REAL*4). The default is 1.010 5.

nzb

- Number of vertical grid levels used for MC-SCIPUFF calculation

(INTEGER*4). No default is available.

zb

- Vertical grid levels used for MC-SCIPUFF calculation (REAL*4).

There must be nzb levels specified. No default is available.

dt_swift

- Update time interval (sec.) for the SWIFT mass-consistent wind

field calculation. (REAL*4). The default is no time binning.

lswift

- Flag indicating availability of SWIFT mass-consistent wind field

module (LOGICAL). Default is 'True'.

178

DRAFT
13.1.5 Sampler Location File.

Time history of concentration at a set of sampler locations can be obtained by

specifying a sampler location file on the input data (Section 13.1.2). In addition to the
mean concentration, the concentration fluctuation variance, and also a fluctuation
correlation time scale are also written to the output file (See Section 13.3.3). The
location file is an ASCII text file containing the concentration variable of interest and a
set of coordinate locations. Only one material with an optional subgroup definition can
be selected for the sampler output. The format for the file is:
matname, igrp : Material name (CHARACTER) and subgroup identifier
(INTEGER*4), free format but separated by white space. The material name must
correspond to a one of the names defined in the MATDEF namelists in the input file
(Section 13.1.2). The subgroup identifier is optional and can be used to select a particular
size group for a particle material, or select vapor (igrp = 1) liquid droplet (igrp = 2) for a

liquid material. A value of zero (default value if igrp is missing) implies total
concentration for the specified material, i.e., sum over all subgroups.
x, y, z : List of sampler locations (one per line) free format. The horizontal

location is in project coordinates; the vertical location is in meters above ground level.
The sample file illustrated in Figure 13-4 defines sampler out put for the vapor
phase of the liquid material GB at 5 locations. Assuming the project is in lat/lon
coordinates, there are 3 samplers on an East-West line at 38N from 105W to 107W all
at 2m above the ground, and 2 more samplers at 4m and 8m height at (105W, 38N).
GB 1
-105.
-106.
-107.
-105.
-105.

38.
38.
38.
38.
38.

2.
2.
2.
4.
8.

Figure 13-4. Example PC-SCIPUFF sampler location file.

179

DRAFT
13.1.6 Radiative Dose Factor File

Radiative dose factors for nuclear facility materials are read in from a project
input file with .RAD extension. As described in Section 8.3.1, human radiation doses are
modeled from combinations of component doses, which are stored on the adaptive
surface grid as the integral of the appropriate dose rates. The RAD file contains the timedependent factors relating the radiation dose rates to the local surrogate material
concentrations. It also contains the definitions of the composite radiation doses, which
are combinations of the component doses, and are available for output display after the
dispersion calculation. Every project material designated as an NFAC material
(including the particle, liquid, aerosol, and gas types NFACP, NFACL, NFACA, and
NFACG, see Section 13.1.2) must have an entry in the RAD file defining the dose rate
factors.
The RAD file consists of a number of keywords, most with associated numerical
or character string input. The general file layout is shown in Figure 13-5. Only
keywords recognized by SCIPUFF are described here; all other keywords are ignored.
The section of the file pertaining to SCIPUFF input is contained between the keywords
BEGIN_SCIPUFF_DATA and END_SCIPUFF_DATA. SCIPUFF will ignore any
information outside the range of these keywords. The order of keywords must generally
follow the order shown in the figure. In particular, material definitions must come first,
followed by component dose definitions, which must precede plotted dose definitions.
Dose rate factors, which may vary with time, come at the end of the SCIPUFF data
section, preceded by the BEGIN_DOSE_FACTORS keyword.
Some general rules regarding the RAD file format are as follows:
1) The maximum length of any line is 128 characters.
2) Comments may be given throughout the file and are denoted by !.
comments are allowed.
3) All blank lines are ignored.
4) Keywords are not case sensitive.

180

In-line

DRAFT
5) Keywords with associated data always have a following = character; data are given
to the right of the equals sign, except for cloudshine distances and dose rate factors,
which are given sequentially on subsequent lines.
6) All data are blank delimited (comma delimited is not valid).
7) As many lines as needed to satisfy data requirements following a keyword may be
used.
8) Character string input, including material names, units and dose names, cannot have
embedded blanks.
9) Each keyword will begin a new line
10) .Units are not checked or used in any SCIPUFF computation; the only use of units
involves the display of plotted doses
Table 13-1 defines the keywords recognized by SCIPUFF and describes the
associated data inputs. Note that the units associated with the EXPOSURE_TIME
keyword are assumed to be days unless defined otherwise with the optional
EXPOSURE_DURATION keyword. Also note the units of the dose rate factors (given
after the EXPOSURE_TIME keyword) are assumed to be dose _ unit sec units m 3
where dose_unit is the component dose unit, and units is the material mass unit specified
in 13.1.2.

Following the EXPOSURE_TIME keyword, dose rate factors are given for each
component dose in the order given by the DOSE_FACTOR keywords. For each
component dose, the number of factors is determined by the number of contributing
materials (AIR or GROUND) multiplied by the number of size groups. Only AIR doses
specified as BY_SIZE use multiple factors corresponding to material size bins. The order
of the materials is defined by the appropriate (AIR or GROUND) MATERIAL keywords.
Remote gamma cloudshine is specified as a component dose by the SHINE descriptor
(only one SHINE dose is permitted), and the dose rate factors are given at each
cloudshine distance location for each AIR dose material. Note that although SHINE dose
is based on AIR dose materials, the rate factors cannot given by size.

181

DRAFT
...
BEGIN_SCIPUFF_DATA
[! comment]
[EXPOSURE_DURATION= tdur [time_unit]]
NUMBER_MATERIALS_AIR= nmatair
MATERIAL= air_mat1 nsize_air1
[size1(1) size1(2) ... size1(nsize_air1)]
MATERIAL= air_mat2 nsize_air2
[size2(1) size2(2) ... size2(nsize_air2)]
...
NUMBER_MATERIALS_GROUND= nmatground
MATERIAL= grnd_mat1 nsize_grnd1
[size1(1) size1(2) size1(3) ... size1(nsize_grnd1)]
...
NUMBER_DOSE_FACTORS= num_dose
DOSE_FACTOR= dose1 dose_unit1 dose_type1
DOSE_FACTOR= dose2 dose_unit2 dose_type2
...
NUMBER_PLOTTED_DOSES= num_plot_dose
PLOTTED_DOSE_NAME= plot_dose1
PLOTTED_DOSE_UNITS= plot_dose_unit1
NUMBER_COMPONENT_DOSES= num_comp_dose1
COMPONENT_NAMES= comp_dose1 factor1
COMPONENT_NAMES= comp_dose2 factor2
...
NUMBER_CLOUD_SHINE_DISTANCES= num_dist
[dist1 dist2 ... distnum_dist]
BEGIN_DOSE_FACTORS
EXPOSURE_TIME= t1 [time_unit]
df1(1) df1(2) ...df1(nmat1)
df2(1) df2(2) ...df2(nmat2)
...
EXPOSURE_TIME= t2 [time_unit]
...
END_DOSE_FACTORS
END_SCIPUFF_DATA
...

Figure 13-5. General structure of a RAD file. Keywords are in upper case; lower case
italics indicate user-specified input. Brackets denote optional input.

182

DRAFT
Table 131.

Description of recognized keywords and associated input for RAD files.

Keyword

Description / Input

BEGIN_SCIPUFF_DATA

Mark beginning of SCIPUFF input data section

END_SCIPUFF_DATA

Mark end of SCIPUFF input data section

EXPOSURE_DURATION

Total exposure time (REAL); not used.

Optional time units can be MIN, HR or D
(days). Default units are days.

NUMBER_MATERIALS_AIR

Number of materials (INTEGER) contributing

to air doses. Must be followed by this number
of MATERIAL keyword inputs.

NUMBER_MATERIALS_GROUND Number of materials (INTEGER) contributing

to doses based on surface deposition. Must be
followed by this number of MATERIAL
keyword inputs.
MATERIAL

Material name (CHARACTER); must

correspond to a NFAC project material. The
number of size groups is given after the name
and must equal the number of subgroups
defined for the project material. The line(s)
following typically contain the mean size group
diameters, but are not used.

NUMBER_DOSE_FACTORS

Total number of component doses (INTEGER,

NC from Section 8.3.1). Must be followed by
this number of DOSE_FACTOR keyword
inputs.

DOSE_FACTOR

The component dose name (CHARACTER),

followed by its units (CHARACTER),and dose
type (CHARACTER),. As discussed in Section
8.3.1, the dose type can be AIR, GROUND,
or SHINE'. AIR component doses may also
be defined according to size group, in which
case AIR is followed by BY_SIZE.

183

DRAFT
Table 131 (contd).
NUMBER_PLOTTED_DOSES

Number of plotted doses (INTEGER) for output

display. Must be followed by this number of
PLOTTED_DOSE_NAME keyword inputs.

PLOTTED_DOSE_NAME

Name of display dose (CHARACTER). Must

be followed by PLOTTED_DOSE_UNITS,
NUMBER_COMPONENT_DOSES and
COMPONENT_NAMES keyword inputs.

PLOTTED_DOSE_UNITS

Units of display dose (CHARACTER).

NUMBER_COMPONENT_DOSES

Number of component doses (INTEGER) used

in computing the plotted dose weighted sum,
see (8.27). Must be followed by this number of
COMPONENT_NAMES keyword inputs.

COMPONENT_NAMES

Component dose name (CHARACTER); must

match a dose defined with the DOSE_FACTOR
keyword. Followed by a multiplicative factor
(REAL), ij in (8.27).

NUMBER_CLOUD_SHINE_
DISTANCES

Number of distances (INTEGER) used to define

the cloudshine function Fcloud of Section 8.3.2.
Must be followed by this number of distances
(meters). If zero, cloudshine is not used.

BEGIN_DOSE_FACTORS

Marks beginning of dose factor input section.

END_DOSE_FACTORS

Marks end of dose factor input section.

EXPOSURE_TIME

Time of dose factors (REAL), units same as for

EXPOSURE_DURATION keyword. Followed
by dose factor data input, corresponding to ij
in Section 8.3.1. Order is determined by the
DOSE_FACTOR keyword input A factor must
be given for each material (and size group or
cloudshine distance if applicable) contributing
to the component dose.

184

DRAFT
13.2 METEOROLOGY INPUT.
13.2.1 Observation File Format.

A meteorological observation input file consists of a header section that specifies

the file type, the number of observation variables, their names and units, followed by the
numerical data.
The header structure is shown in Figure 13-6 and individual records are described
in the following.
comment : Optional comments are indicated by # in the first column. There is
no limit to the number of comment lines.
keyword : For PROFILE files (see ftype below), comment lines containing a

colon are searched for recognized keywords. Keywords typically have associated input
read in from the string to the right of the colon. Recognized keywords and corresponding
input strings are given in Table 13-2. The TYPE and ANALYSIS keywords characterize
the observational data for use in the meteorology uncertainty calculation; see Section
12.4. (The ANALYSIS keyword is used only for FORECAST input.) The
TIMEREFERENCE keyword specifies the time convention of the input file.
ftype : The file type which can be 'PROFILE' or 'SURFACE'. (Only these two
types are recognized by SCIPUFF.) A SURFACE file has only one observation for each
station and is typically associated with near-surface measurements of wind and/or
boundary layer parameters such as surface heat flux or mixing-layer height. A PROFILE
file generally has more than one observation height at a station and is used when vertical

profiles of wind, temperature, etc. are available, e.g. from upper air soundings.
nvar : The number of observation variables given in the file, read with free
format. If nvarp is given, then nvar is the number of fixed variables for a PROFILE file.

Fixed variables are not functions of height and these are indicated in Table 13-3. nvar
must be at least 4 (station id, time and horizontal position are the minimum
requirements). nvar + nvarp cannot exceed 30 for a PROFILE file; nvar cannot exceed
22 for a surface file.

185

DRAFT
[# comment]
...
[# keyword: string]
...
ftype
nvar [nvarp]
name1 name2 name3 ... namenvar
unit1 unit2 unit3 ... unitnvar
[pname1 pname2 pname3 ... pnamenvarp]
[punit1 punit2 punit3 ... punitnvar]
missing [zref]

Figure 13-6. Header structure for a meteorology observation file. Records and variables
in brackets are optional or required only if nvarp is given.

Table 132.

Valid keywords and associated input strings for PROFILE files.

Keyword

String

Description

TYPE

OBSERVATION (default)

Characterization of the data source

as observations, analyzed fields or
a numerical forecast.

ANALYSIS
FORECAST
ANALYSIS

YEAR, MONTH, DAY,

HOUR (free format)

Analysis time used to initialize the

numerical forecast

TIMEREFERENCE

LOCAL

Time convention for the

meteorology file

UTC (default)

nvarp : The number of profile observation variables, read with free format. This

is an option for PROFILE files only. Profile variables typically vary with height.
Examples are height, pressure, temperature, humidity and wind velocity. The minimum
number is 3 (height and velocity).

186

DRAFT
name1, ... , namenvar : The observation variable names are read using (30a8)
format. There must be nvar names specified and the order here determines the order in
which variables are given in each record of the numerical data. Names recognized in
SCIPUFF are given in Table 13-3; multiple specific names indicate synonyms. Other
names will be ignored without causing an error so that input files with variables not

needed in SCIPUFF may still be used. If nvarp is given, these are the names of the fixed
variables. Acceptable fixed variable names are indicated in Table 13-3. Not all
combinations of recognized variable names are permitted. This is discussed below.
unit1, ... , unitnvar : The observation variable units are read using (30a8) format.

Only certain variables require that units be specified; otherwise blank spaces are
acceptable. Variables requiring units are indicated in Table 13-3 along with those units
recognized by SCIPUFF. An error message is given if a unit is unrecognized (but only
for those requiring units).
pname1, ... , pnamenvarp : The profile observation variable names are read as

described for name, if nvarp is given. There must be nvarp names specified.
punit1, ... , punitnvarp : The profile observation variable units are read as
described for unit, if nvarp is given.
missing : A character string of up to 8 characters used to indicate a missing or
bad value in the numerical data.
zref :

Optionally, the reference height in meters above the surface can be

specified for SURFACE observations. If the reference height is omitted, a standard value
of 10m is assumed, unless the height for each observation is specified in the data record.
The minimum requirements for meteorological observation files are that the
station ID, time, location, wind speed and height (for a PROFILE file only) must be
specified. Additionally, if certain input options are chosen, such as observational largescale variability (Section 12.3) or observational or profile boundary layer then specific
combinations of variables must be given. In particular, then, the following variables (or
some combination thereof) are required as a minimum:
1) ID,
2) time (TIME, HOUR, YEAR, MONTH, DAY, YYMMDD, JDAY),
3) wind velocity (U, V, WSPD, SPEED, SPD, DIR,WDIR),

187

DRAFT
4) horizontal location (X, Y, LON, LAT),
5) Z if file type is PROFILE,
6) UL, VL and UVL if observational large-scale variability is specified
(ENSM_TYPE=OBS; see Section 13.1.4),
7) Both ZI and HFLUX together or PGT or MOL if observational boundary layer is
specified (BL_TYPE=OBS; see Section 13.1.4),
8) UU, VV, WW, WT, SL, and SZ if profile boundary layer is specified
(BL_TYPE=PROF; see Section 13.1.4).
Some combinations of variables, particularly those concerning time, are not
allowed, are mutually exclusive or result in some of the variables being ignored. Also,
the presence or absence of certain variables dictates that others be given. These
situations are now enumerated:
1) Specifying TIME supersedes HOUR, DAY, MONTH, YEAR, YYMMDD and
JDAY. No error message is given, but all other time variables are ignored.
2) HOUR must be specified in the absence of TIME. HOUR must be combined with
either YYMMDD, the combination of DAY, MONTH, YEAR or JDAY. Note that
YYMMDD supersedes DAY, MONTH, YEAR, which supersedes JDAY. All three
variables DAY, MONTH and YEAR must be given together.
3) X and Y must be specified together. X and Y supersede LON and LAT.
4) LON and LAT must be specified together and are required if X and Y are not given.
5) U and V must be specified together. U and V supersede WSPD, etc. and DIR or
WDIR.
6) WSPD, etc. and DIR or WDIR must be specified together and are required in the
absence of U and V.
7) SIGU, SIGV and RHOUV or UL, VL and UVL must be specified together. These
can only be used in combination with U and V; an error occurs if they are used with
WSPD and WDIR. SIGU, SIGV and RHOUV are only used if the hazard area is
being calculated (Sections 9.4 and 12.4); otherwise they are ignored.
8) MOL supersedes PGT and HFLUX; PGT supersedes HFLUX.

188

DRAFT
Table 133. Meteorological variables and their units recognized in SCIPUFF.
Variable Description

Time
(intended for laboratory or other
idealized situations where date and
time of day is irrelevant)

Units

ID

1,2

Station ID

Specific Name

None

TIME

HOURS
HRS
MIN
SEC

HOUR

HOURS

YEAR
MONTH

None

DAY

None

YYMMDD

None

JDAY

None

KM

KM

Year

LON

None

Time of Day
2
2

Month
2

Day

Year-Month-Day
Julian Day
X-location
Y-location
Longitude

3
3

W
2

Latitude

LAT

S
Wind Speed

WSPD or

KNOTS

SPEED or

KTS

SPD

M/S
MPH

Wind Direction

DIR or
WDIR

189

DEG

DRAFT
Table 133 (Continued). Meteorological variables and their units recognized in
SCIPUFF.
Variable Description

Specific Name

Units

Velocity along X- or LongitudeAxis

KNOTS
KTS
M/S
MPH

Velocity along Y- or Latitude-Axis

KNOTS
KTS
M/S
MPH

Height

M
FEET
FT
3

Pressure

MB

Temperature

C
K
F

Humidity

GM/GM

HUMID

G/G

GM/KG
G/KG
%

Boundary Layer Mixing Height

ZI

M
FEET
FT

Pasquill-Gifford-Turner Stability
2
Class

PGT

190

None

DRAFT
Table 133 (Continued). Meteorological variables and their units recognized in
SCIPUFF.
Variable Description

Specific Name
HFLUX

Surface Heat Flux

Units
K-M/S
C-M/S
W/M/M
W/M2

Monin-Obukov Length

MOL

2,6

PRCP

None

SIGU

KNOTS

Precipitation Index

U-Velocity Std. Deviation, Met.

Uncertainty

KTS
M/S
MPH

V-Velocity Std. Deviation, Met.

Uncertainty

SIGV

KNOTS
KTS
M/S
MPH

Met. Uncertainty Velocity

Correlation Coefficient

RHOUV

None

U-Velocity Std. Deviation, LargeScale Variability

UL

KNOTS
KTS
M/S
MPH

V-Velocity Std. Deviation, LargeScale Variability

VL

KNOTS
KTS
M/S
MPH

Large-Scale Variability Velocity

Correlation Coefficient

UVL

191

None

DRAFT
Table 133 (Continued). Meteorological variables and their units recognized in
SCIPUFF.
Variable Description

Specific Name

Units

Shear-Driven Lateral Velocity

Variance

UU

M2/S2

Buoyancy-Driven Lateral Velocity

Variance

VV

M2/S2

Vertical Velocity Variance

WW

M2/S2

Boundary Layer Heat Flux

WT

K-M/S
C-M/S

Turbulence Scale (Buoyancy-Driven)

SL

Turbulence Scale (Shear-Driven)

SZ

Notes:
1
ID can be an 8 character alphanumeric string if it is the first variable given,
otherwise it must be an integer.
2
variable can be specified as fixed data on a PROFILE file.
3
unit assumed regardless of input.
4
default unit if none specified.
5
standard PGT classes converted to numerical values, i.e. class A = 1, B = 2, ...,
G = 7 (non-integer values are permitted).
6
precipitation index specified as follows: 0=none, 1=light rain, 2=moderate
rain, 3=heavy rain, 4=light snow, 5=moderate snow and 6=heavy snow.
The numerical data are given after the header section in the order dictated by the
list of variable names. The general structure of the numerical data is shown in Figure
13-7. If nvarp is given, individual profiles must be separated by the fixed data record,
which must begin with the string ID:. Vertical profiles must be given with height
monotonically increasing. This is also true for time and it is further necessary that all
observations at a particular time be grouped together. All nvar and nvarp, if applicable,
variables for each station and/or height at a particular time must be contained on a single
line of no more than 256 characters. The figure illustrates that all nz heights for station
id1 at time1 are given before station id2 and that data from all nsta stations at time1 are
given before time2.

192

DRAFT
var1 ... id1

... time1 ... z1

... varnvar

(a)

...
var1 ... id1
... time1 ... znz ... varnvar
var1 ... id2
... time1 ... z1 ... varnvar
...
var1 ... idnsta ... time1 ... znz ... varnvar
var1 ... id1
... time2 ... z1 ... varnvar
...
ID: id1... time1 ... varnvar
var1 ... z1 ... varnvarp
...
var1 ... znz ... varnvarp
ID: id2... time1 ... varnvar

(b)

var1 ... z1 ... varnvarp

...
var1 ... znz ... varnvarp
...
ID: idnsta... time1 ... varnvar
var1 ... z1 ... varnvarp
...
var1 ... znz ... varnvarp
...
ID: id1... time2... varnvar
var1 ... z1 ... varnvarp
...
var1 ... znz ... varnvarp
...

Figure 13-7. General structure of the numerical data in a meteorological observation file.
(a) Without nvarp specified; (b) with nvar fixed variables and nvarp
profile variables.

193

DRAFT
13.2.2 MEDOC Format.

A MEDOC file contains a header section followed by numerical data; multiple

times require a complete header and data sections for each time for each time. A sample
Fortran code that reads the data for a single time is illustrated in Figure 13-8. It should be
noted that the MEDOC format was designed for other applications and therefore includes
variables that are not used as SCIPUFF input. These are indicated as dummy variables in
the figure. Also, some extensions to the original MEDOC format have been made, such
as the staggered grid and spherical coordinate system options.
The first record must be either 'FFFFFFFF' for a formatted file or 'BBBBBBBB'
for a binary file. The binary file is read with code identical to that shown in Figure 13 8,
absent the format statements. The definitions of the input variables for the rest of the
records are as follows:

CODENAME : Name of code used to generate data. Currently ignored in

SCIPUFF.
STAGGER : Flag indicating if grid is staggered, i.e., the velocity components
are shifted half a grid length from the center of the grid cell along their respective
coordinate directions. If the first character of STAGGER is 'T', a staggered grid is

assumed. Otherwise, or if a value for STAGGER is not given, an unstaggered grid is

assumed
IDAY, IMONTH, IYEAR, IHOUR, IMIN, ISEC : Day, month, year, hour,

minute and seconds defining time of following data.

IMAX, JMAX, KMAX : Number of grid points in the (x, y, z) directions,
respectively.
NREPER : Number of special points. This information is not used for the
SCIPUFF input and NREPER can be set to zero.
NVAR3D : Number of three-dimensional fields.
NAM3D : Names of the three-dimensional variables. SCIPUFF requires at least
the 2 horizontal velocity component fields, with the names 'U', 'V'. SCIPUFF will also
read the vertical velocity component, 'W' (required if terrain elevation is given as a twodimensional field), potential temperature 'T' or absolute temperature 'TA', humidity ratio,

194

DRAFT
'H', and the meteorological uncertainty variance fields 'UUE', 'VVE' and the
corresponding correlation coefficient 'UVE'. Velocities are assumed to be in ms1,
temperature in K, humidity in grams moisture per grams dry air, and variances in m2s2.
All other names are ignored.

C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C

C
C
C
C
C
21
C
C

RECORD 1 - FILE FORMAT ('FFFFFFFF' or 'BBBBBBBB')

READ (1,9001) FFLAG
RECORD 2 - NAME OF CODE AND STAGGERED GRID FLAG
READ (1,9001) CODENAME, STAGGER
RECORD 3 - TIME
READ (1,9002) IDAY,IMONTH,IYEAR,IHOUR,IMIN,ISEC
RECORD 4 - INITIAL TIME OF CALCULATION - NOT USED IN SCIPUFF
READ (1,9002) JDAY,JMONTH,JYEAR,JHOUR,JMIN,JSEC
RECORD 5 - NUMBER OF GRID POINTS, KEY POINTS, VARIABLES
READ (1,9002) IMAX,JMAX,KMAX,NREPER,NVAR3D,NVAR2D
RECORD 6 - NOT USED FOR SCIPUFF
READ (1,9002) IDUM,IDUM,IDUM,IDUM,IDUM,IDUM
RECORD 7 - NOT USED FOR SCIPUFF
READ (1,9002) IDUM,IDUM,IDUM
RECORD 8 - GRID AND TIMING INFORMATION (KMAX+11 VALUES)
READ (1,9003) (SZ(K),K=1,KMAX),DX,DY,XO,YO,LAT,LON,
&
DUM,DUM,DUM,DUM,ZTOP
RECORD 9 - NAMES AND UNITS (NREPER+2*NVAR3D+2*NVAR2D VALUES)
READ (1,9001) (NAMDUM,N=1,NREPER),
&
(NAM3D(N),N=1,NVAR3D),(NAMDUM,N=1,NVAR3D),
&
(NAM2D(N),N=1,NVAR2D),(NAMDUM,N=1,NVAR2D)
RECORD 10 - NOT USED IN SCIPUFF
READ (1,9003) (DUM,N=1,3*NREPER)
RECORD 11 - 3D VARIABLES
(NVAR3D SETS OF IMAX*JMAX*KMAX VALUES)
DO 21 N=1,NVAR3D
READ (1,9003) (((VAR3D(I,J,K,N),I=1,IMAX),J=1,JMAX),K=1,KMAX)
RECORD 12 - 2D VARIABLES (NVAR2D SETS OF IMAX*JMAX VALUES)
DO 41 N=1,NVAR2D
READ (1,9003) ((VAR2D(I,J,N),I=1,IMAX),J=1,JMAX)

41
C
9001 FORMAT(6(A8,1X))
9002 FORMAT(6(I12,1X))
9003 FORMAT(6(F12.4,1X))

Figure 13-8. Fortran pseudo-code for reading a formatted MEDOC input file.

195

DRAFT
NVAR2D : Number of two-dimensional fields.
NAM2D : Names of the two-dimensional variables. SCIPUFF recognizes the
following two-dimensional fields: terrain elevation in meters (required if vertical velocity

is given), 'REL' or 'TOPO'; planetary boundary layer height (above surface) in meters, 'ZI'
or 'PBL_HITE'; surface heat flux in watts per square meter, 'HFLX' or 'SFC_HTFX';
roughness height in meters, 'ZRUF'; surface albedo, 'ALBEDO'; Bowen ratio, 'BOWEN';
canopy height, 'CANOPY'; and canopy flow index, 'ALPHA'. All other names are
ignored.
SZ : Array of KMAX vertical grid coordinates in meters. The MEDOC terrain-

following coordinate transformation is described in Section 10.3.2.

DX, DY : Horizontal grid spacings in meters or degrees, depending on the

coordinate system.
XO, YO : Horizontal grid origin (SW corner) coordinates in km for Cartesian
coordinate system; values of 999999 indicate a spherical coordinate system
(latitude/longitude).
LAT, LON : Horizontal grid origin (SW corner) coordinates in degrees.
ZTOP : Vertical meteorology domain height in meters. If the parameter is not

specified , the domain height is assumed to be SZ(KMAX). For a staggered grid, the
value of ZTOP indicates which of two vertical grids is used. If ZTOP = SZ(KMAX), a
SWIFT grid is assumed, where the first level of the vertical velocity component is at
SZ(KMAX) 2. If ZTOP = [ SZ(KMAX) SZ(KMAX) ] 2, a MC-SCIPUFF grid is
assumed, where the first level of the vertical velocity component is zero (at the surface).

13.2.3 HPAC Format.

An HPAC file contains a header section specifying grid parameters and the
number of variables, followed by the numerical data. The general format structure is
illustrated in Figure 13-9.
The header variables in the first record are read with a (4f10.0, 5i5) format and
are defined as follows:
xlon0, xlat0 : Origin of the longitude/latitude grid (lower left corner) in degrees

E and N, respectively.
196

DRAFT
xlon0, xlat0, dx, dy, nx, ny, nz, nt, nvar
z1, z2, ... znz
p1, p2, ... pnz
(month, day, year, time)1
title1
var1((((i,j,k),i=1,nx),j=1,ny),k=1,nz)
...
titlenvar
varnvar((((i,j,k),i=1,nx),j=1,ny),k=1,nz)
(month, day, year, time)2
title1
var1((((i,j,k),i=1,nx),j=1,ny),k=1,nz)
...
titlenvar
varnvar((((i,j,k),i=1,nx),j=1,ny),k=1,nz)
(month, day, year, time)nt
...

Figure 13-9. General structure of an HPAC gridded meteorology file

dx, dy : Grid spacing in degrees of longitude and latitude, respectively.

nx, ny :

Number of grid points in the longitude and latitude directions,

respectively.
nz : Number of vertical grid levels.
nt : Number of time breaks.
nvar : Number of variables. Currently, only nvar = 4 is valid. The field

variables are assumed to be, in order, u-wind component (east-west, ms1), v-wind
component (north-south, ms1), temperature (C) and relative humidity (%). If nvar is
missing, a default value of 4 is assumed.

197

DRAFT
The next record(s) gives the nz vertical grid levels, read using (10f8.0) format.
The following record(s) contains the corresponding pressures and is also read with
(10f8.0) format.
The numerical data follows the first three records (or 3 + 2*int[(nz 1)/10] if nz
is greater than 10). As illustrated in Figure 13-9, each time break is headed by a record
giving month, day, year, hour, and minutes, read with (1x, i2, 1x, i2, 1x, i2, ix, i2, i2)
format. Figure 13-9 also shows that a title record, read with (a12) format, precedes the
data for each field variable. This record normally contains the field variable name and
units, but SCIPUFF does not check consistency between this title and the assumed
variables and order described above. The field data itself is read using (10f8.0) format so
that the number of records for each field is int[(nx*ny*nz 1)/10] + 1. The indexing
order of the fields is shown in Figure 13-9.

13.2.4 Terrain File Format.

The terrain input file is an ASCII text file containing a grid description header
followed by terrain elevation data defined on a regular two dimensional grid. The
header consists of an optional record followed by two mandatory records. The general
structure of the terrain file is illustrated in Figure 13-10.

The first record is required only if a landcover description is given on the file. In
that case, the keyword LAND_USE followed by an equals sign must begin the record.
(Any other keyword will result in an error.) The names to the right of the equals sign
denote the landcover fields on the file. The permissible names and their meanings are:
ZRUF, surface roughness; ALBED, surface albedo; BOWEN, Bowen ratio; HCANP,
canopy height; ALPHC, canopy flow index; and CATEGORY, landcover category index.
Any combination of the first three fields is allowed, but the canopy height and flow index
must be given together. The landcover category is given in lieu of the first five fields.
Valid category values and the corresponding landcover descriptions are given in Table
13-4.
The rest of the header records are defined as follows:
grid_type, param1, param2 : the grid type (CHARACTER*8) can be
LATLON, UTM, KM, METERS or M. If grid_type is LATLON, the two

198

DRAFT
numerical parameters are not required. For UTM grids, param1 (INTEGER) contains
the UTM zone number; the other parameter is not required. For all other types, if
param1 (REAL) and param2 (REAL) are specified, they contain the longitude and
latitude of the grid origin, respectively. The numerical parameters are in free format.
grid_type, param1, param2 : the grid type (CHARACTER*8) can be
LATLON, UTM, KM, METERS or M. If grid_type is LATLON, the two
numerical parameters are not required. For UTM grids, param1 (INTEGER) contains
the UTM zone number; the other parameter is not required. For all other types, if
param1 (REAL) and param2 (REAL) are specified, they contain the longitude and

latitude of the grid origin, respectively. The numerical parameters are in free format.
x0, y0, dx, dy, nx, ny : the grid origin (REAL) for the east-west coordinate (x0),

the grid origin (REAL) for the north-south coordinate (y0), grid spacings (REAL) (dx
and dy) and number of grid points in the coordinate directions (INTEGER) (nx and ny).
x0 and y0 are in degrees if grid_type is LATLON, otherwise they are in kilometers. dx
and dy are in the same units as the origin coordinates unless grid_type is METERS or M
in which case they are in meters.
Hi,j : the terrain height in meters read with (12i6) format.

LAND_USE = land_key1

land_key2

land_keyN

grid_type, param1, param2

x0, y0, dx, dy, nx, ny
H1,1, H2,1,...,Hnx,1, H1,2,..., Hnx,ny
land_key1
L11,1, L12,1,...,L1nx,1, L11,2,..., L1nx,ny
...
land_keyN
LN1,1, LN2,1,...,LNnx,1, LN1,2,..., LNnx,ny

Figure 13-10. General structure of a terrain file.

199

DRAFT
Table 134.

Valid landuse indices and associated landuse types for terrain files.
Landuse Index

Landuse Type

Developed

Dry Cropland & Pasture

Irrigated Cropland

Missing

Cropland/Grassland

Cropland/Woodland

Grassland

Shrubland

Shrubland/Grassland

10

Savanna

11

Deciduous Broadleaf Forest

12

Deciduous Needleleaf Forest

13

Evergreen Broadleaf Forest

14

Evergreen Needleleaf Forest

15

Mixed Forest

16

Water

17

Herbaceous Wetland

18

Wooded Wetland

19

Barren

20

Herbaceous Tundra

21

Wooded Tundra

22

Mixed Tundra

23

Bare Tundra

24

Snow or Ice

25

Partly Developed

200

DRAFT
The following records are given only if the first record contains the keyword
LAND_USE, as described above
Land_KeyN : the name of the landcover field to follow, which must match one

of the names given on the first record.

L1i,j , ,LNi,j : the landcover fields corresponding to land_key1 , ,
land_keyN, read with (12f10.0) format for all fields except CATEGORY, which is
read with (12i6) format.

The location (i, j) is taken to be at x0+(i-1)*dx and y0+(j-1)*dy.

Note that fields in both UTM and spherical coordinate systems can be included in
a single terrain file essentially by combining two separate files. The UTM fields must
come first and conform to the record description just given, including the header
information. The spherical coordinate system records then follow, starting with the
header records. The coordinate system chosen for the puff calculation will determine
which fields are read.

13.3 OUTPUT FILES.

13.3.1 Surface Output Files.

The surface integrals of dose and deposition, described in Section 2, are stored as
adaptive grid files with multiple timebreaks. The dose integral, T , is stored in the
surface file, project.dos, and the deposition, D , in project.dep. These files are directaccess binary files with a record length of 512 bytes (128 words), and consist of a number
of timebreaks. Each timebreak contains a header record followed by the field data; files
created by HPAC 3.2 (PC-SCIPUFF version 1.3) contain additional file header records
and a modified timebreak header.

The file header for HPAC 3.2 is written using the Fortran statements in Figure
13-11. In order to accommodate the generalized radiation dose computations described

201

DRAFT
in Section 8, the new format uses a "block" description, in which several output fields
associated with a particular material or effect are grouped together. The standard dosage
or deposition block for any material contains 3 fields, namely the ensemble mean, the
variance, and the correlation scale for the specified variable. The block is described by
an overall block name, composed of the material name and additional descriptors for
liquids and particles. Thus for a gas or aerosol material, the block name is simply the
material name. For a particle material, PART, the block names are 'Total PART' for the
total over all size groups and 'PART Size group 3' if the block contains only size group 3.
For a liquid material, LIQD, the block name can be 'Total LIQD', 'Vapor LIQD', or
'Liquid LIQD'; in addition, for a liquid with secondary evaporation, the block contains 2
extra variables defining the mass-weighted droplet size and effective wetted area.
Special blocks are stored for radiation doses, and the architecture has been designed for
future expansion to describe other types of effects. The explicit names for the radiation
dose blocks are determined by the Nuclear Facilities source module. Each block also
contains a pointer into the total list of variables on the surface file, indicating the start of
the block. The file header uses the following variables:
file_type

(character*13)

must be set to 'BLOCK_VERSION'

nblk

(integer*4)

number of data blocks on file

iversion

(integer*4)

PC-SCIPUFF version number

blk_name

(character*64)

Array of data block names

ifield

(integer*4)

Array of pointers to indicate start of blocks

nvar

(integer*4)

Total number of variables in file

names

(character*4)

Individual field names, e.g. "Mean", "Var"

c----- FILE HEADER RECORDS

........ir = 1
write(nwrt,rec=ir) file_type,nblk,iversion
do i = 1,nblk
ir = ir + 1
write(nwrt,rec=ir) blk_name(i),ifield(i)
........end do
ir = ir + 1
write(nwrt,rec=ir) nvar,(names(i),i=1,nvar)

Figure 13-11. Fortran statements to write a surface output file header (HPAC3.2 only).

202

DRAFT
The Fortran statements to write a timebreak are illustrated in Figure 13-12.

The timebreak header record consists of the variables:

time

(real*4)

time of the field in hours from the beginning of the

calculation

ngrid

(integer*4)

total number of grid cells in the adaptive grid. A value of

zero indicates the end of the file.

(nx, ny)

(integer*4)

number of base grid points in the (x, y)-direction

(xorg, yorg) (real*4)

origin coordinates of the base grid

(dx, dy)

(real*4)

grid spacings of the base grid

nvar

(integer*4)

number of field variables

names

(char*4)

array of field variable identifying names

iversion

(integer*4)

PC-SCIPUFF version number

c----- TIMEBREAK HEADER RECORD

if (file_type .eq. 'BLOCK_VERSION) then
write(nwrt,rec=ir) time, ngrid, nx, ny, xorg, yorg, dx, dy,
$
nvar, iversion
else
write(nwrt,rec=ir) time, ngrid, nx, ny, xorg, yorg, dx, dy,
$
nvar, (names(i),i=1,nvar), iversion
end if
C----- GRID INDEX ARRAY
do ii = 1,ngrid,128
j1 = min0(ii+127,ngrid)
ir = ir + 1
write(nwrt,rec=ir) (iref(j), j=ii,j1)
end do
C----- FIELD DATA ARRAYS
do ivar = 1,nvar
do ii = 1,ngrid,128
j1 = min0(ii+127,ngrid)
ir = ir + 1
write(nwrt,rec=ir) (fields(j,ivar), j=ii,j1)
end do
end do

Figure 13-12. Fortran statements to write a surface output file timebreak.

203

DRAFT
Note that the new format (BLOCK_VERSION) does not place the variable names
on the timebreak header record. For the old format, the variable names are constructed
from the input material list but do not contain direct reference to the material identifier
names. The field names are 4-character identifiers, and the first character denotes the
type of field. M denotes the mean field, V denotes the fluctuation variance, and S
denotes the correlation length scale. Note that the scale variable is multiplied by the
fluctuation variance on the output file, since the integral value is weighted by the
variance. Units for the various fields are dependent on the type of surface field, but are
based on the material mass units (as specified as part of the material properties
definitions) and SI units. HASCAL runs contain the correlation time scale instead of the
length scale, and also contain D and X fields, which are the mean and variance of the
integrated surface deposition, as defined in Section 2.4. For gaseous materials or
individual particle size bins, the remaining three characters of the name contain the
unique type descriptor number, as defined in Section 13.4. For particle materials with
multiple size bins, the total particulate mass variables are represented by a two-character
integer, containing the material identifier number, and the last character of the name is T.
The material identifier is simply the position in the list of MATDEF namelists in the
project.inp file, as described in Section 13.1. Thus, the total mean deposition for the first
material in the input file would be named as M01T, while the variance of the fifth
particle size bin for the first material would be V005.
The adaptive grid structure is described by the integer*4 index array, iref, and the
field data is stored in the real*4 array fields. For each grid cell, iref points to the first
refinement sub-cell, if it exists. If the grid cell, i, is not refined, then iref(i) = 0. The first
nx*ny cells are the base grid, so ngrid is necessarily greater than or equal to nx*ny. The
centers of these cells are located at
xi = xorg + (i 0.5)*dx,

i = 1, nx

yj = yorg + (j 0.5)*dy,

j = 1, ny

and the location (xi, yj) corresponds to cell number (j 1)*nx + i in the adaptive grid.
If a cell is refined, then the iref array contains the grid cell number of the first of
four sub-cells with half the grid size of the parent cell. Suppose i0 is the parent cell
number, with grid sizes x0 and y0 and center location (x0, y0). The parent cell may be a
base grid cell but is not necessarily so since the refinement process can be continued
without limit. The four sub-cells are numbered consecutively as {i1 + k : k = 0, 3} where
204

DRAFT
iref(i0) = i1. The sub-cells are numbered as shown in Figure 3-3 (i.e., lower left is zero,
then lower right, upper left, upper right), and have center locations

(x0 x 0

4,y 0 y0 4)

and grid size (x0/2, y0/2).

Each cell has a complete set of field variables associated with it, so the local field
value has contributions from all levels of refinement. A simple estimate of the field value
at a given location (xp, yp) can be obtained using the pseudo-code illustrated in Figure
13-13. Note that this example uses a recursive subroutine call to scan the list of
refinements. The field array in the example represents any one of the nvar variables in
the fields array on the file. This estimate simply adds contributions from successive
refinement levels without any interpolation across grid cells.

xx = (xp-xorg)/dx
yy = (yp-yorg)/dy
i0 = int(yy)*nx + int(xx) + 1
cval = 0.0
call sum_point_val(i0,xx,yy,cval)
c ---- cval contains the local field value
c----------------------------------------------------------------------subroutine sum_point_val(i0,xc,yc,cval)
cval = cval + field(i0)
c ***

Check for refinement

if (iref(i0) .ne. 0) then
xc = 2.*(xc - int(xc))
yc = 2.*(yc - int(yc))
i1 = iref(i0) + 2*int(yc) + int(xc)
call sum_point_val(i1,xc,yc,cval)
end if
return
end

Figure 13-13. Fortran statements to calculate field values from the surface output file.

205

DRAFT
13.3.2 Puff File.

The puff file, project.puf, is a binary file containing the complete puff data at a
number of time breaks together with boundary layer and continuous source information
for restart purposes. The file is written using the unformatted Fortran write statement
write(lun_puf) t,
$

npuf,((p_real(i,j),j=1,36),(p_int(i,j),j=1,7),i=1,npuf),

npaux,(paux(i),i=1,npaux-1),

nspecies,

nxbl,nybl,dxbl,dybl,

(xbl(i),i=1,nxbl),(ybl(i),i=1,nybl),

(zibl(i),i=1,nxbl*nybl),

ncrel,(caux(i),i=1,ncrel)

nsrcaux,(srcaux(i),i=1,nsrcaux)

where
t

(real*4)

time of the data in seconds from the beginning of the

calculation

npuf

(integer*4)

total number of puffs

p_real(i,*)

(real*4)

floating point data for puff i

p_int(i,*)

(integer*4)

integer data for puff i

npaux

(integer*4)

total number of puff auxiliary data values + 1

paux(*)

(integer*4)

floating point data auxiliary puff data

nspecies

(integer*4)

integer reserved for implementation of multicomponent

chemistry, currently set to 0

nxbl,nybl

(integer*4)

horizontal dimensions of meteorology grid

dxbl,dybl

(real*4)

horizontal grid intervals of meteorology grid

xbl(*),ybl(*) (real*4)

horizontal grid locations of meteorology grid

zibl(*)

(real*4)

boundary layer depth on meteorology grid

ncrel

(integer*4)

number of active continuous sources

caux(*)

(real*4)

auxiliary data for restarting continuous sources

206

DRAFT
nsrcaux

(integer*4)

total number of auxiliary variables for restarting

continuous pool sources

srcaux(*)

(real*4)

auxiliary data for restarting pool sources

The floating point puff data is stored as follows:

j=1-3

x, y, z

puff centroid coordinates; see (2.3)

j=4-9

ij

puff spread moments, in the order xx, xy, xz, yy, yz,

zz; see (2.4)

j=10-15

ij

puff inverse moments, in the order xx, xy, xz, yy, yz,
zz; see (2.62)

j=16

determinant of the puff spread moments

j=17

puff mass; see (2.2)

j=18

c2

j=19

x u c

integrated mean square concentration; see (2.54)

L

Large-scale diagonal-xx turbulent flux moment; see

Section 2.1.3

j=20

X12

j=21

y v c

j=22

y v c

j=23

y v c

cross-xy turbulent flux moment; see (2.14)

L
B
S

Large-scale diagonal-yy turbulent flux moment

Buoyancy-driven diagonal-yy turbulent flux moment
Shear-driven diagonal-yy turbulent flux moment; see
Section 2.1.3

j=24

z w c

diagonal-zz turbulent flux moment; see Section 2.1.3

j=25

w c

vertical turbulent drift moment; see (2.33)

j=26

c2

squared mean concentration from overlap calculation; see

(2.58)

j=27-28

c , cH

horizontal concentration fluctuation length scales; see

Section 2.2.2

j=29

cV

vertical concentration fluctuation length scale; see

Section 2.2.2

j=30

2Tc

correlation time scale for the puff concentration

fluctuations; see (2.97)

j=31

fA

activity factor; see Section 8.1

j=32

zi

local mixed layer depth; see Section 10.2.4

207

DRAFT
j=33

zc

puff vertical cap height; used to control puff splitting

within the planetary boundary layer, see Section 3.3

j=34-36

u, v, w

ambient wind velocity for use with Adams-Bashforth

scheme; see Section 3.4

The integer puff data is stored as follows:

j=1

type

an identifier for the puff material and size bin type

j=2-3

next, prev

pointers for the linked list on the spatial grid

j=4

grid

grid refinement level for the spatial grid

j=5

tlev

refinement level for the puff time step

j=6

tnxt

pointers for the linked list for time stepping

j=7

iaux

pointer into the auxiliary array for additional puff data.

The number of data values in the array for any given puff
in dependent on the material type and on the overall
dynamics and dense_gas flags

There are a possible 25 additional puff data values for each puff. These are
j=1

2
cT

mean square total concentration (for multiple size bin

particle materials); see (2.49)

j=2

c T2

squared mean total concentration (for multiple size bin

particle materials); see (2.53)

j=3-10

buoyancy parameters

j=11-21

dense gas dynamics parameters

j=22-25

liquid material parameters

The puff type identifier is simply allocated sequentially in the order that materials
are specified in the project.inp file. Thus, the first material is type = 1 if it is a gas, or the
Nbin size bins are type = 1 up to type = Nbin for a particle material and .type = Nbin +1 for
a liquid material to include the vapor phase. Identifiers for the second material are then
allocated beginning with the next available value for type.

208

DRAFT
13.3.3 Sampler Time-History File.

The sampler time history file is an ASCII text file containing the instantaneous
concentration field statistics for the selected material at the locations specified in the
sampler location file described in Section 13.1.5. The output file consists of a header
section followed by a single data record for each model time step. The output interval is
the large time step as specified in the input file. The header records are

Record 1: The number of variables per data record (3 per sample location + 1)
Record 2: A list of variable names(CHARACTER*8)
Record 3: A title string containing the project name
The data records contain the time (in seconds from the start of the calculation) and
three data values for each sampler location:

Data Record: time, c , c 2 , Tc for location-1, c , c 2 , Tc for location-2, etc.

where the three data values are mean concentration, concentration variance, and
correlation time scale (in seconds), respectively.
An example of the sampler output file is given in Figure 13-14. The output shows
3 sampler locations, implying 10 variables per record. The output time interval is 10
seconds and the data records only show the first 5 of the 10 output variables.

7
T
C001
V001
T001
C002
Sampler time histories for GB vapor for
10.00000 0.0000000E+00 0.0000000E+00
20.00000 0.0000000E+00 0.0000000E+00
30.00000 1.0000000E-03 1.2000000E-05
40.00000 1.5000000E-03 2.0000000E-05
50.00000 2.0000000E-03 3.0000000E-05
60.00000 4.0000000E-03 4.0000000E-05
.....
.....

Figure 13-14. Example of a sampler output file.

209

V002
T002
project gbrel
0.0000000E+00
0.0000000E+00
1.0000000E+00
1.2000000E+00
1.8000000E+00
2.0000000E+00

0.0000000E+00...
0.0000000E+00...
0.0000000E+00...
0.0000000E+00...
0.0000000E+00...
0.0000000E+00...

DRAFT
SECTION 14
MODEL TEST CALCULATIONS
14.1 SHORT RANGE DIFFUSION.
14.1.1 Pasquill-Gifford-Turner Stability Categories.

The commonly used Pasquill-Gifford-Turner (PGT) stability curves for plume

dispersion are valid for short-range dispersion from surface releases and are based on an
averaging time of about 10 mins. The PGT dispersion estimates are empirical fits to
atmospheric observations, and are therefore representative of real data. The PGT
classification system requires a self-similar solution to the diffusion equation, so that the
plume spread is a function of downstream distance and stability class only. This is
possible if the turbulent velocity scales are proportional to the transport speed and the
turbulent length scales are only dependent on the vertical height. The standard surface
layer representation fulfills most of these conditions, since the turbulence and mean wind
speed are generally specified in terms of u* and dimensionless height z/L, where L is the
Monin-Obukhov length.
We note that the buoyancy component requires knowledge of the mixed layer
depth, and we assume a fixed value of 1000m for these comparisons. The MoninObukov length corresponding to each stability category is given Table 11-2. The
contributions from the shear and buoyancy components are scaled as described in Section
12.1. With this specification for the atmospheric turbulence, we can compute an almost
self-similar solution for the vertical dispersion using the closure model, but the horizontal
dispersion is more complicated. We use a release at zrel = 5m with a surface roughness
length of 3cm and a wind speed of 5ms1 at z = 10m, although it should be noted that the
results are dependent on all these choices. The SCIPUFF predictions for horizontal and
vertical plume spread are shown in Figure 14-1, and stable conditions are omitted since
the horizontal turbulence generation mechanisms in a stable stratification are poorly
understood. The closure model is in reasonable agreement with the generally accepted
dispersion curves for surface layer dispersion, although we note the dependence on wind
speed and other input choices.

210

DRAFT

100

y (m)
10

1
F

100

z (m)
10
A
1
F
1

10

100

1000

x (m)
Figure 14-1. Comparison between SCIPUFF predictions (solid lines) and the PGT
dispersion curves for categories A through F.

14.1.2 Model Data Archive.

The Model Data Archive (MDA), compiled by Sigma Research Corporation, is an

extensive collection of field experiment data for both dense and passive tracer gas
releases. The data set contains both continuous and instantaneous sources with different
sampling times, and has been used to evaluate a number of existing atmospheric
dispersion models (Hanna et al., 1991). Using the model coefficients defined from the
211

DRAFT
preliminary test calculations, SCIPUFF has been compared with the test cases in the
Model Data Archive (MDA). The experimental data are divided into a passive release set
and a dense gas release set but we deal only with the passive cases here; the dense gas
comparisons are described in Section 14.4.
The passive releases are relatively straightforward, since the only source
parameters are the mass flow rate and the source size. The initial standard deviation of
the puffs is set equal to the source radius for these releases, which include the Prairie
Grass and Hanford data. The meteorology for the dispersion calculation was specified as
a single wind speed at the appropriate observation height, together with a surface heat
flux and a boundary layer depth, zi. The Prairie Grass data set includes an estimate of the
Monin-Obukhov length, L, and the surface friction velocity, which directly provide a heat
flux. The Hanford data includes a vertical temperature difference, which was used in
conjunction with the wind speed to determine the surface heat flux. Measurements of the
boundary layer depth were not available, so zi was set to 1000m for neutral and unstable
cases, and to 5L for the stable cases.
The data specifies an averaging time for the concentration data, and both short
and long duration measurements are included. The observed values represent the
maximum recorded, over all sampling locations at a given distance and over the sampling
period. For the long duration measurements, which usually correspond to the release
duration, the model is run using the conditional dispersion mode, described in Section
2.2.5. The concentration sampling period, Tavg, is the appropriate averaging time for
filtering the turbulence, and the centerline maximum prediction at the sampler
downstream distance is then used for comparison.
For the short duration data, we use the statistical SCIPUFF prediction to estimate
the expected value. When a short duration peak, say a 1-second average, is reported from
a number of samplers over a sampling period, Tsample, it must be recognized that this is a
random variable. For an idealized continuous release of infinite duration, we expect the
maximum observed value to increase as the sampling period increases, since there is a
higher probability of measuring an extreme value. In order to model the expected peak
value, we use the predicted probability distribution for the centerline concentration, c0,
using a turbulence filter of Tavg. If only one measurement (in time) of the maximum
concentration over a number of sampler locations across the plume is made, then the

212

DRAFT
expected value would be c0 . However, if we sample for a longer time, then the expected
value must increase since the chances of encountering a higher value are increased.
Suppose that the probability density function for c0 is such that
Prob c 0 < X = P0 X

a f

Then, if the maximum from N independent observations of c0 is denoted as cN, then

Prob c N < X = PN X

a f
P a Xf

since all N observations must be less than X. If we can estimate N, therefore, we can
determine the expected value of cN as the predicted value for the observed maximum
concentration. N is estimated from the reduction in the concentration fluctuation variance
with averaging time (Sykes, 1984). If we identify this reduction with the factor 1/N for
the variance reduction from averaging N independent samples of a discrete variable, then
we obtain
N 1 =

d 1 + e i

(13.1)

where = Tsample Tc and Tc is the integral timescale of the fluctuations.

The calculation procedure, therefore, is:
(1) reduce c 2 to account for the averaging time, Tavg,
(2) determine the clipped normal distribution P0(X) for the reduced variance,
(3) determine the exponent N from (13.1),
(4) calculate the expected maximum value from the distribution PN.
This procedure is applied to the sampler time series to obtain the predicted
maximum value.
Results for the overall comparison are shown in Figure 14-2, for the passive data
sets. The overall agreement with the range of data is relatively good. Most of the
predictions are within a factor of 2 of the observations, and there is reasonable skill in
predicting the effect of time averaging on the maximum concentration.

213

DRAFT

Predicted Concentration (ppm)

103

101

101

a - Prairie Grass
b - Hanford (Long time average)
c - Hanford (Short time average)

a
a

a a

aa
a a aa
a
aa
b
a a
aa a
a
a
aa a
a c
a aa ca
a a a a ba
a
a
ab c
b

a
a
a
aaa a a
a
a
a
a aa a
aa
a aa a
a
a a aaa a
a a a aaaa a a
aa
a
a
a
aa
a a a aa a
ac
aa aa a
a
a
aa aa a
b
a aaa aa
a aa
aa aaa
aa a a
a
a
aaa
a a aaaa a
a a aa
c
aa
a
c a a aaaaa a
b ac
aaa aa
a a aa aa
a
b a aaa a
aa
aa a a ca a
c b
aa
a
a aaa b
a
aa aa a
ba
c
aa a a
a

101

101

103

Observed Concentration (ppm)

Figure 14-2. Comparison between the observed maximum concentrations and SCIPUFF
predictions for the passive releases in the Model Data Archive. Various
experiments are indicated by the letter codes.

14.1.3 Instantaneous Dispersion.

Several short-range instantaneous dispersion experiments have been compared
with SCIPUFF predictions. These comparisons are presented in the paper by Sykes and
Gabruk (1997) and are only briefly described here.
The first data comparison uses the ice crystal experiment results of Weil et al.
(1993). A cluster of ice crystals was formed within a natural cloud by seeding with dry-

214

DRAFT
ice pellets from an aircraft, and the evolution of the ice crystal cloud was then measured
from the aircraft along with relevant turbulence statistics. Weil et al. present averaged
results for the dimensionless cloud spread from an ensemble of about 25 releases, and the
data are compared with the closure model prediction in Figure 14-3. We use an initial
source size of 53m, which corresponds to the larger of the source sizes considered by
Weil et al., and the data have been made dimensional using the observed average values,

v 2 = 3.6m2s2 and = 0.013m2s3. We assume isotropic conditions to determine the

other velocity variances, and the turbulence length scales are obtained from the
dissipation rate (see Lewellen, 1977) as
S = V =

bq 3

(13.2)

which implies S = 341m for the Weil et al. data.

10000

1000

x (m)
100

10
10

100

1000

10000

t (sec)
Figure 14-3. Comparison between SCIPUFF predictions and the observational data of
Weil et al. (1993).

215

DRAFT
Figure 14-4 shows the SCIPUFF comparison with the data of Mikkelsen et al.
(1988) for smoke dispersion in atmospheric surface layer. The lateral turbulent velocity
variance, v 2 , was measured as 0.98m2s2, and the Lagrangian integral time scale was
estimated to be 100s from the measured Eulerian spectrum. The mean velocity at the
plume height was 4.72ms1. We again assume isotropic conditions for simplicity since
the other turbulence components were not reported, although we note that this
assumption is not strictly valid in the atmospheric surface layer. The length scale is
derived from the Lagrangian time scale, which corresponds to /Aq in the closure model,
and the vertical and horizontal scales are assumed to be equal. The initial source size is
taken to be 0.5m, corresponding to the larger value tested by Mikkelsen et al. The results
are not sensitive to the assumption of isotropy, since we use the observed value of the
Lagrangian time scale to determine the length scale. A test calculation using S = 6.5m,
as appropriate for a plume at z = 10m, and reducing the vertical velocity variance in

1000

100

y (m)
10

0.1
1

10

100

1000

t (sec)
Figure 14-4. Comparison between SCIPUFF predictions and the observational data of
Mikkelsen et al. (1988).

216

DRAFT
accord with the 5/3 law, gave almost identical results to the isotropic calculation, as
shown in the figure.
Figure 14-5 compares the closure model prediction with the neutral relative
dispersion data of Hgstrm (1964), as described by Sawford (1982). The release was
made at 50m and there is considerable uncertainty with regard to the turbulence
conditions. We follow Sawford in determining the effective length scale from the latetime dispersion measurements. We use a turbulence intensity of 0.1, as suggested by
Sawford, and again assume isotropic conditions. Thus, u 2 = v 2 = w 2 = 0.01U 2 , and
the length scales, S and V, were taken to be 16m. A source size of 1m was used in the
calculation, since this is the midrange value considered by Sawford. The turbulence
specifications enforce agreement with the data at late time, but the model does predict the
proper early time growth of the plume.

1000

100

z (m)

10

1
10

100

1000

10000

t (sec)

Figure 14-5. Comparison between SCIPUFF predictions and the observational data of
Hgstrm (1964).

217

DRAFT
14.1.4 CONFLUX.
SCIPUFF simulations were compared with data from the Concentration
Fluctuation Experiments (CONFLUX) project of 1994 (Biltoft, 1995). Figures 14-6 and
14-7 show the vertical profiles of mean concentration and standard deviation to mean
concentration ratio of the CONFLUX data and SCIPUFF simulations corresponding to
the first six trials examined in Yee et al. (1995).
In this field study, concentration time series were constructed from measurements
of neutrally buoyant plumes in slightly convective to moderately stable meteorological
conditions. All concentration data were obtained with a vertical array of 16
photoionization detectors at downwind locations between 12.5 and 100m.
Contemporaneous measurements of wind velocity and temperature supplied the
meteorological input needed to perform SCIPUFF simulations. We assumed isotropic
conditions and partitioned the horizontal velocity variance into shear and buoyant
components as described in Section 12.1, with length scales, B and S, of 250m and
0.65z, respectively. The shear component was determined from the observed shear
stress, using (12.3), and the remainder of the horizontal energy was assigned to the
buoyancy-driven component. A source size of 0.05m with a release heights of from 1 to
5m was used in the calculations. The sampling period of the concentration measurements
was approximately 35 minutes, which was sufficiently long to sample most of the
turbulent energy spectrum. Profiles of mean concentration and concentration variance
were calculated by time averaging the data. SCIPUFF simulations were conducted in the
conditional dispersion mode using a concentration averaging time, Tavg, equal to the
measurement sampling period.
The comparisons indicate reasonably good agreement on the vertical dispersion
rate, and concentration levels. Note that the scales are linear, so that most of the
predictions are closer than a factor of 2. The concentration fluctuation intensity is also in
good agreement with the observations, and high values are observed. This is due to the
small source size, which produces a narrow plume that is meandered by the large
atmospheric eddies. Both the magnitude and shape of the distribution is predicted
accurately.

218

DRAFT
16
CC03

CC05

CC06

CC07

CC08

CC09

12
Z
8
(m)
4
0
16
12
Z
8
(m)
4
0
16
12
Z
8
(m)
4
0

8
12
C (ppm)

16

8 12
C (ppm)

16

Figure 14-6. Comparison between SCIPUFF predictions of mean concentration and the
observational data from CONFLUX 1994. The trial designations are taken
from Yee et al. (1995).

219

DRAFT
16
CC03

CC05

CC06

CC07

CC08

CC09

12
Z 8
(m)
4
0
16
12
Z
8
(m)
4
0
16
12
Z 8
(m)
4

12 16 20
c/C

0
0

12 16 20
c/C

Figure 14-7. Comparison between SCIPUFF predictions of standard deviation to mean

concentration ratio and the observational data of CONFLUX 1994.

220

DRAFT
14.2 LONG RANGE DIFFUSION.
SCIPUFF has been tested against two major long-range atmospheric dispersion
experiments.
First, the Across North America Tracer Experiment (ANATEX)
comparisons were described by Sykes et al. (1993c). The published results were obtained
using an earlier version of SCIPUFF, but the dispersion algorithms are effectively the
same. A more sophisticated treatment of the different turbulence populations from the
boundary layer and mesoscale has been implemented, as described in Section 2.1.3, but
the longer range predictions are dominated by the mesoscale component and are not
sensitive to the combination methodology.
The second experiment is the more recent European Tracer Experiment (ETEX)
(Mosca et al., 1997). Figure 14-8 shows a comparison between a SCIPUFF simulation
and ETEX data at several times after the start of the release. In this study, an inert tracer
was released for 12 hours at an average rate of 7.95 gs1. The evolution of the tracer
cloud was measured for a period of 72 hours (from the beginning of the release) with a
sampling network of 168 sites. The meteorological input to SCIPUFF was the
uninitialized, gridded analysis supplied by the European Centre of Medium Range
Weather Forecasts (ECMWF). Additionally, turbulence input for the second-order
closure model was specified as two separate populations, representing the planetary
boundary layer eddies and the larger scale contribution. Boundary layer turbulence
profiles were determined using the diurnal boundary layer calculation described in
Section 11.2 with no cloud cover. We used a Bowen ratio of 0.6, and a surface albedo of
0.16; these values are typical of mid-latitude rural terrain, but clearly cannot represent the
detailed variations over the dispersion domain. The large-scale turbulence represents the
mesoscale fluctuations, and is based on the work of Gifford (1988).
The comparison with the observations shows that the location and shape of the
cloud is reasonably well predicted as it travels across the continent, and the magnitude of
the predicted surface concentration is also in reasonable agreement with the observed
values. Figure 14-9 shows the measured concentration at a number of sampler locations
and also the SCIPUFF prediction. The concentrations are averaged over 3 hour periods,
and the time is shown relative to the start of the release. The agreement in arrival time
and peak concentration magnitude is reasonably good.

221

DRAFT
SCIPUFF
24 Hours

Latitude (degrees)

24 Hours

48 Hours

48 Hours

72 Hours

64

Observed

72 Hours

59
54
49
44

Latitude (degrees)

64
59
54
49
44

Latitude (degrees)

64
59
54
49
44
-5

10

15

20

25 -5

10

15

20

25

Longitude (degrees)

Longitude (degrees)

Figure 14-8. Comparison of mean concentration between SCIPUFF predictions and

ETEX data (Mosca et al., 1997). Contours are 0.01, 0.02, 0.05, 0.1, 0.2, 0.5,
1, 2, 5, 10, 20, and 50 ngm3.

222

DRAFT
2.0

B5

NL1

DK5

DK2

D42

D5

CR3

PL3

H2

1.5

1.0

.5
0
2.0
1.5

c
1.0

.5
0
2.0
1.5

1.0

.5

20

40

60

80

20

40 60
t (hrs)

80

20

40

60

80

Figure 14-9. Comparison between 3-hr average concentrations from SCIPUFF

predictions (dashed) and ETEX data (solid line) at 9 selected locations
(Mosca et al., 1997). Concentration is in ngm3.

223

DRAFT
The 9 stations were chosen from the two sampler arcs analyzed by Mosca et al.,
showing passage of the plume at roughly 24 hours and 48 hours. Only 2 samplers from
the middle of the first arc are shown; there are 2 other samplers on this arc but they
generally showed the poorest model performance. The reason for this is apparently that
the actual plume moved further to the North than predicted by the given wind field. Most
of the models underpredict the Northern station and overpredict the Southern station, and
we note that the SCIPUFF results are consistent with this trend although they are not
shown here.
The results shown in the Figures 14-9 are not identical to the SCIPUFF
predictions reported by Mosca et al. (1997), since an older version of the model was used
there. Two of the reported overall measures were the fractional bias and normalized
mean square error (NMSE). These measures are defined as

( Pi M i )

2
N

M +P

( Pi M i )

1
NMSE =
N

FB =

MP

where
M i and P =

1
N

1
N

M=

Pi
i

are the mean measured and predicted values, respectively. N is the number of
observations, and the subscript i denotes individual samplers. The SCIPUFF prediction
gives a mean prediction for all 168 sampling locations of 0.096 ngm3, while the mean
observation is 0.054 ngm3, giving a fractional bias of 0.57 and a NMSE of 228.
The difficulty with these measures is the lack of normalization of the
concentrations, so that errors in the large values can dominate the statistics. In the
SCIPUFF predictions, the sampler location closest to the release completely determines
these two statistics, since it contains the highest observations and SCIPUFF overpredicts
the values. The meteorological data provided for the calculation also lacked a boundary
layer description, so the early plume development is strongly influenced by the model
choice for the boundary layer. Table 14-1 shows the model performance statistics for
different time periods. The early times are clearly responsible for most of the overprediction with model performance improving significantly after the first 15 hours. The

224

DRAFT
bias for sampling times between 30 and 75 hours is almost negligible and NMSE is also
relatively low. During the last 15-hour period, SCIPUFF shows an under-prediction and
increased NMSE, although the value of 22.4 is not as large as the early time values.
Figure 14-10 shows a comparison between the time-integrated dosage at each of
the sampling locations. The first panel shows direct comparisons between the predicted
and observed values, and indicates a relatively low fraction within a factor of 2 as
indicated by the dashed lines. The discrepancies are largely due to errors in the cloud
trajectory, as indicated by the second panel, which shows ordered pairs. Here, the lists
of observed and predicted dosages are simply re-ordered by magnitude, so that the largest
observation is plotted against the largest prediction and so on down the list. The good
agreement in this comparison shows that, in general, the predicted dosages correctly
cover the range of observed doses but occur at different sampling locations. An
exception is visible in the largest dosages, which show the model over-prediction at 3
locations. As discussed above, these locations are closest to the source and are most
influenced by uncertainties in the wind field and boundary layer definition.

Time period

FB

NMSE

No. of
observations

All times

0.57

228

2743

0 15 hrs

1.80

1836

269

15 30 hrs

0.82

46.5

560

30 45 hrs

0.068

14.8

593

45 60 hrs

0.045

5.51

564

60 75 hrs

0.067

8.36

510

75 90 hrs

0.48

22.4

247

Table 14-1. Model performance statistics for SCIPUFF prediction of the ETEX sampler
concentrations for different time periods.

225

DRAFT
6

10

(a)

10

++ +
+
++ +
+ +
+
+++ ++ +
+
++
+ ++++++
++ +
+ ++ ++ + +
+ +
+ + + ++ ++ ++
+ + +
+
+
+ ++ + +
++ + + + + + + + +
+ +
+
+ +
+ + ++ + + +
+
++ +
++
+ +
+
++
++
+
+
++
+
+
++
+
+
+
+

Predicted
Dose
4

10

10

10

10
10

10

10

10
Observed Dose

10

10

10

+
+
+

(b)
5

10

+
+
+
+
+
+
+
++
++
++
++
+++
+++
+
+
+
+
+
++
++
+
++
+
+
++
++
++
+
+
++
++
+
++
++
++
+
+ ++
+
+ +
+
+

Predicted
Dose
4

10

10

10

10

10

10

10

10
Observed Dose

10

10

Figure 14-10. Comparison between total integrated dosage from SCIPUFF predictions
and ETEX data (solid line) at all locations (Mosca et al., 1997). Dosage is
in ng s m3. (a) compares each station, (b) shows ordered data, unmatched
by location.

226

DRAFT
14.3 DYNAMIC RISE EFFECTS.
The vertical rise dynamics in SCIPUFF has been validated principally against
idealized laboratory experiments or in comparison with well-accepted empirical relations
based on laboratory and field experiments. These include momentum jets in a crossflow,
buoyant plumes in neutral or stable temperature gradients with a crossflow and buoyant
spherical bubbles released into a neutrally-stable quiescent background.

14.3.1 Momentum Jet.

The first experiments considered are momentum jets (no buoyancy) in a

crossflow. Gordier (1959) (from Hirst, 1971) presents near-source data for a variety of
exit velocity ratios, R = W0 U a , where Ua is the crossflow velocity and W0 is the jet exit
velocity. The jet source in SCIPUFF is characterized by a momentum flux Fm and
Gaussian spread 0, so that W0 is estimated by assuming an initially uniform velocity
distribution such that W02 = Fm r02 where r0 is the source radius. [Fm is related to the
SCIPUFF momentum flux Qm by Fm = Qm ]. We assume that 0 = r0 . A comparison
with the mean jet centerline data of Gordier is shown in Figure 14-11. SCIPUFF tends to
under-predict the jet heights; this is particularly so near the source, but the discrepancies
tend to decrease with downstream distance. The high R data show much less downstream
displacement of the jet close to source; this is probably due to distortion of the
background flow, a phenomenon not included in the SCIPUFF model.
The asymptotic behavior of the momentum jet far downstream is also of interest
and here we compare SCIPUFF with the well-known expression (see Weil, 1988 or
Briggs, 1975)

13

(13.3)

13

where x is downstream distance, z is the jet centerline height above the source, is an
empirical entrainment constant typically set to 0.6 and the momentum length scale
m = Rr0 . Figure 14-12 shows the SCIPUFF calculations for three exit velocity ratios
compared with (13.3). It can be seen that, while there is naturally some initial dependence
on R, the three cases show the same asymptotic behavior. The model predictions are
consistent with the one-third power law but indicate a value for around 0.5.

227

DRAFT

70
R
2.5
4.0
6.0
8.0
13.2
15.4

60

50

R=15.4
13.2

z r0
40

8.0
6.0

30

4.0

20

2.5
10

0
0

Figure 14-11.

10

20

30

x r0

40

50

60

70

Momentum (non-buoyant) jet centerline height, z / r0 , as a function of

downstream distance for a range of R. Symbols are the data of Gordier
(1959) - from Hirst (1971), solid lines are the SCIPUFF predictions.

228

DRAFT
10 2

10 1

R=15.4

SCIPUFF
"One-third" law = 0.5
"One-third" law = 0.6

R=8.0

Figure 14-12.

R=2.5
10 1

10 0
10 0

10 2

10 3

Comparison between SCIPUFF predictions of momentum jet centerline

heights for a range of R (solid lines) and the one-third law, (13.3), with
two values of (short dashes, = 0.6; long dashes, = 0.5).

14.3.2 Buoyant Plume.

For a buoyancy-dominated plume in a neutral background, the plume height is

given by the two-thirds power law (Briggs, 1975):

23

(13.4)

2 2

13

3
where the buoyancy length scale is given by b = Fb U a . The buoyancy flux, Fb, is
defined as Fb = W0 r02 g T0 Ta where T0 is the source temperature perturbation relative
to the ambient temperature Ta. [Fb is related to the SCIPUFF buoyancy flux QT by
Fb = QT g Ta ]. The two-thirds law has been shown to compare well with numerous
field and laboratory data (Briggs, 1975). The one-third and two-thirds laws are

229

DRAFT
essentially limiting cases of a more general expression. Thus, for plumes with significant
buoyancy and momentum fluxes, the trajectory is given by

 

    
b

m
b

2 2

2 13

(13.5)



which reduces to (13.4) for x >> 2 b and to (13.3) with x << 2 b . A comparison of
m
m
(13.5) with the SCIPUFF predictions for a range of buoyancy fluxes and source sizes is
shown in Figure 14-13. As expected, there are some discrepancies near the source, but the
agreement with the two-thirds law is excellent further downstream.
Plume rise in a uniformly stratified environment (N2 = a positive constant) has
been investigated by Briggs (1975) who gives the following expression, which accounts
for the added mass due to the displacement of ambient fluid by the plume:

10 3

10 2

10 1
B
SCIPUFF
Theory

C
10 0
10 0

10 1

10 2

10 3

10 4

Figure 14-13. Comparison between SCIPUFF predictions for buoyancy-dominated

jet centerline heights (solid lines) and theory given by (13.5) (dashed
4
lines). Case A: R = 10, Fb2 U a Fm = 0.01; Case B: R = 10,
4
4
Fb2 U a Fm = 1 ; Case C: R = 1, Fb2 U a Fm = 1 .

230

DRAFT
13

F 3 IJ FG N F sin N x + 1 cos N x IJ
=G
U K
c F U N h H K H F U
z

2 13

13

(13.6)

and kv is the added mass

for N x U a where N = N 1 + k v , = 1 + k v
coefficient (set to 1, the value for a circular cylinder). Based on an extensive survey of
field and laboratory data, Briggs (1975) also determined an empirical expression for the
final plume rise in stable air as

zfinal = 2.6 Fb U a N 2

12

12

13

(13.7)

SCIPUFF predictions for a range of buoyancy fluxes and N compare quite favorably with
(13.6) and (13.7), as shown in Figure 14-14.

101

Fb Ua N 2

13

A
100

C
SCIPUFF
Theory
Final Rise
10-1
100

10-1

Figure 14-14.

Nx
Ua

101

Comparison of SCIPUFF buoyancy-dominated jet centerline heights

in a uniformly stratified background (solid lines) with (13.6) (dashed
4
lines). For all cases, R = 10. Case A: Fb2 U a Fm = 0.01,
4
2
4
4
Fb N U a = 0.00572 ; Case B: Fb U a Fm = 1 , Fb N U a = 0.00572 ;
Case C: Fb2 U a4 Fm = 1 , Fb N U a4 = 0.181.

231

DRAFT

14.3.3 Buoyant Puff.

The plume rise relations just presented are based on extensive laboratory and field
experiments and are universally accepted, at least under the idealized conditions
assumed. In contrast, the paucity of data for short duration or nearly instantaneous
buoyant releases precludes any generally accepted simple formulae. The only data we
currently can compare SCIPUFF with come from experiments on the rise of light density
spherical bubbles conducted in the early 1970s under DNA sponsorship (Mantrom and
Haigh, 1973). These tests involved the release of helium bubbles in air or N2 bubbles in
SF6 for a range of Reynolds numbers. The backgrounds were neutrally stratified and
quiescent. For large enough Reynolds numbers, the bubble rise data generally collapse to
a single curve of z D0 vs. t g D0 where, z is the bubble centroid height above its
release point, D0 is the initial mean bubble diameter and t is time after release. The
SCIPUFF predictions are compared with the data in Figure 14-15. Two releases are
considered for the model calculations, the first using a single spherical puff with an initial
Gaussian spread equal to the bubble radius while the second uses 136 puffs in an attempt
to model explicitly an initial spherical bubble with a uniform mass distribution. It can be
seen that the agreement with the data is generally quite good for both releases, although
the multi-puff calculation tends to under-predict the bubble rise slightly. The late time
behavior in both the data and model calculations seems consistent with a power law
exponent slightly less than two-thirds.

14.4 DENSE GAS EFFECTS.

14.4.1 Model Data Archive.

The Model Data Archive (MDA), discussed in Section 14.1.2, also contains a
dense gas release set comprised of several experiments. The dense releases include
evaporating pools (Burro, Coyote, and Maplin Sands experiments), continuous plumes
and instantaneous clouds of heavy gas (Thorney Island) and two-phase aerosol jets
(Desert Tortoise and Goldfish).

232

DRAFT
6
Re
13000
10400

z
D0

B
4

(a)

10

12

14

16

18

20

g
t
D0
102
Re
13,000
10,400
27,600
101

z
D0

100

(b)
10-1 -1
10

100

g
t
D0

101

102

Figure 14-15. Centroid height of a light bubble released into a neutral quiescent
background. Symbols are the data of Mantrom and Haigh (1973), solid
lines are the SCIPUFF predictions. The Reynolds number for the
experiments are based on the initial bubble diameter and the terminal
velocity of a corresponding non-entraining sphere. Case A is for a single
puff release with a Gaussian spread of D0/2; Case B is for a release
comprised of 136 puffs representing a uniform distribution of mass within a
spherical bubble of diameter D0. (a) Early time behavior; (b) expanded scale
showing late time behavior.

233

DRAFT
The Thorney Island tests used Freon gas and involved no evaporative effects, so
the mass and mass fluxes were used directly for the continuous and instantaneous
releases respectively. The standard deviations were set to 75% of the specified source
diameter.
For evaporating pools, we use a lateral standard deviation equal to 75% of the
specified pool radius and a vertical standard deviation of 0.5m for all cases. The vertical
size is chosen arbitrarily to represent a typical vertical growth during transport across the
pool, and results are not sensitive to the precise value. The initial buoyancy flux was
determined from the given mass flux and the difference between the gas boiling
temperature and the ambient value. We assume that the vaporizing material is released at
the boiling point, so that the buoyancy flux is
FB =

Q Tb Ta c (pgas )

air c

( air )
p

(13.8)

where cp is the specific heat, Tb is the boiling point, Ta is the air temperature, and Q is the
mass flow rate.
The Desert Tortoise and Goldfish releases involve a two-phase aerosol jet and
exothermic reaction with the environment. Since SCIPUFF does not contain a model for
aerosol evaporation, and also the thermodynamics of the reactions were not prescribed,
any model result will depend significantly on the choice for initial buoyancy conditions.
We use a simplified representation, modeling the source jet as an expanding 7 cone until
the sufficient air has been entrained to evaporate the liquid aerosol. The SCIPUFF source
is then initialized at this downstream location at the boiling temperature. The initial size
is chosen to match the area implied by the expanding cone but with a horizontal spread
that is twice the vertical. This procedure is similar to that used by Hanna et al. (1991) for
other models that do not address the phase change effects.
The meteorology was specified in the same way as for the passive releases, except
that no surface heat flux information was available for the Thorney Island continuous
releases. For these cases, the estimate of the Monin-Obukhov length provided by Hanna
et al. (1991) as input for the SLAB model was used.

234

DRAFT
The effects of time averaging were accounted for as described in 14.1.2, and the
predicted centerline maxima are compared with the MDA observations in Figure 14-16.
Agreement with the observations is comparable with the passive releases shown in Figure
14-2.

Predicted Concentration (ppm)

106

105

a - Burro (long)
b - Burro (short)
c - Coyote (long)
d - Coyote (short)
e - Maplin (propylene)
f - Maplin (LNG)

104

103

102
102

b b a
b
h
b h
aa h f
h
h hh bfd
h ea
i
h a c bef ab ge
ba a
b
eb bjd ied a
h
jfib df g
eh e
e hhjh h dh ih
ee j f h
c a
a d fh f e g
ea
h aed ch e g
c aa
fc e d b f egh
d
e
c e ge
b
h d d f bh e
hfc ec ahh e g
h
h he ah he
ef
fee
h
c
eh
e
gf
c
e h akh
h e b h bfe
e
c e h h
h
i
a
ki
h hee e ea
k
h f h h h gi e
g
j
gje h j
i
e h
h g e jh e
h
h g he
kh h
g
g hh h
k
k

k
k

g - Thorney Island (continuous)

h - Thorney Island (instantaneous)
i - Desert Tortoise (short)
j - Desert Tortoise (long)
k - Goldfish

103
104
105
Observed Concentration (ppm)

106

Figure 14-16. Comparison between the observed maximum concentrations and SCIPUFF
predictions for the dense gas releases in the Model Data Archive. Various
experiments are indicated by the letter codes.

235

DRAFT
SECTION 15
REFERENCES
Anno, G. H., G. E. McClellan, and M. A. Dore (1996), "Protracted radiation-induced
performance decrement: Volume 1 Model Development", Defense
Nuclear Agency, DNA-TR-95-117-V1 (U).
Athey, G. F., A. L. Sjoreen, J. V. Ramsdell, and T. J. McKenna (1993), "RASCAL
Version 2.0 User's Guide", NUREG/CR-5247, Vol 1, Rev. 1 (PNL-8454),
U.S. Nuclear Regulatory Commission, Washington, D.C. (U).
Baines, W. D. and D. F. James (1994), "Evaporation of a droplet on a surface", Ind. &
Eng. Chem. Res., 33, 411-416 (U).
Bass, A. (1980), "Modelling long range transport and diffusion", Second Joint Conf. on
Appl. of Air Poll. Met., AMS & APCA, New Orleans, LA, pp193-215
(U).
Biltoft, C. A. (1995), "Surface effects on concentration fluctuation profiles", Joint
Contact Point Directorate, Dugway Proving Ground, DPG/JCP-95019 (U).
Boris, J. (1986), "A vectorized 'near neighbors' algorithm of order N using a monotonic
logical grid", J. Comp. Phys., 66, 1-20 (U).
Brown, A. R., S. H. Derbyshire and P. J. Mason (1994), "Large-eddy simulation of stable
atmospheric boundary layers with a revised stochastic subgrid model",
Quart. J. Roy. Met. Soc., 120, 1485-1512 (U).
Briggs, G. A. (1975), "Plume rise predictions", Lectures on Air Pollution and
Environmental Impact Analysis, 59-111, ed. D. A. Haugen,
Am. Met. Soc., Boston MA (U).
Businger, J. A. (1973), "Turbulent transfer in the atmospheric surface layer", Workshop
on Micrometeorology, ed. D. A. Haugen, American Meteorological
Society, Boston, MA (U).
Businger, J. A. (1986), "Evaluation of the accuracy with which dry deposition can be
measured with current micrometeorological techniques",
J. Clim. & Appl. Met., 25, 1100-1124 (U).

236

DRAFT
Businger, J. A., J. C. Wyngaard, Y. Izumi and E. F. Bradley (1971), "Flux-profile
relationships in the atmospheric surface layer",
J. Atmos. Sci., 28, 181-189 (U).
Carson, D. J. (1973), "The development of a dry inversion-capped convectively unstable
boundary layer", Quart. J. Roy. Met. Soc., 99, 450-467 (U).
Carson, D. J. and P. J. R. Richards (1978), "Modelling surface turbulent fluxes in stable
conditions", Boundary-Layer Met., 14, 67-81.
Chamberlain, A. C. (1967), "Transport of Lycopodium spores and other small particles to
rough surfaces", Proc. Roy. Soc., 296A, 45-70 (U).
Cionco, R. M. (1972), "A wind-profile index for canopy flow",
Boundary-Layer Met., 3, 255-263 (U).
Clark, T. L. (1977), "A small-scale dynamic model using a terrain-following coordinate
transformation", J. Comp. Phys., 24, 186-215 (U).
Davidson, C. I., J. M. Miller and M. A. Pleskow (1982), "The influence of surface
structure on predicted particle dry deposition to natural grass canopies",
Water Air Soil Pollut., 18, 25-43 (U).
Deardorff, J. W. (1970), "Convective velocity and temperature scales for the unstable
planetary boundary layer and for Rayleigh convection.",
J. Atmos. Sci., 27, 1211-1213 (U).
Deardorff, J. W. (1980), "Progress in understanding entrainment at the top of a mixed
layer", Workshop on the Planetary Boundary Layer, Amer. Meteor. Soc.,
36-66 (U).
Donaldson, C. du P. (1973), "Atmospheric turbulence and the dispersal of atmospheric
pollutants", AMS Workshop on Micrometeorology, ed. D. A. Haugen,
Science Press, Boston, pp313-390 (U).
Dyer, A. J. (1974), "A review of flux-profile relationships",
Boundary-Layer Met., 7, 363-372 (U).
Egan, B. A. and J. R. Mahoney (1972), "Numerical modeling of advection and diffusion
of urban area source pollutants", J. Appl. Met., 11, 312-322 (U).

237

DRAFT
Ermak, D. L. (1990), "User's manual for SLAB: An atmospheric dispersion model for
denser-than-air releases", Lawrence Livermore National Laboratory,
UCRL-MA-105607 (U).
Etling, D. (1990), "On plume meandering under stable stratification",
Atmos. Env., 24A, 1979-1985 (U).
Fackrell, J. E. and A. G. Robins (1982), "Concentration fluctuations and fluxes in plumes
from a point source in a turbulent boundary layer",
J. Fluid Mech., 117, 1-26 (U).
Friedlander, S. K. (1977), "Smoke, Dust and Haze", Wiley, New York, 317pp (U).
Fuchs, N. A. (1964), "The Mechanics of Aerosols", Pergamon, New York (U).
Gifford, F. A. (1959), "Statistical properties of a fluctuating plume dispersal model",
Adv. Geophys., 6, 117-137 (U).
Gifford, F. A. (1988), "A similarity theory of the tropospheric turbulence energy
spectrum", J. Atmos. Sci., 45, 1370-1379 (U).
Gordier, R. L. (1959), "Studies on fluid jets discharging normally into moving liquid", St.
Anthony Falls Hyd. Lab., University of Minnesota, 28, Series B (U).
Hanna, S. R. (1983), "Lateral turbulence intensity and plume meandering during stable
conditions", J. Clim. Appl. Met., 22, 1424-1431 (U).
Hanna, S. R., D. G. Strimaitis and J. C. Chang (1991), "Hazard response modeling
uncertainty (a quantitative method). Volume II. Evaluation of commonlyused hazardous gas dispersion models", Air Force Engineering and
Service Center (U).
Henn, D. S., R. S. Gabruk, S. F. Parker and R. I. Sykes (1995), "CLOUDTRANS: A
TASS and DIAL-A-CLOUD Output Translator Code", Defense Nuclear
Agency, DNA-TR-94-85 (U).
Hicks, B. B. (1976), "Wind profile relationships from the Wangara experiment", Quart.
J. Roy. Met. Soc., 102, 535-551.
Hirst, E. A. (1971), "Analysis of round, turbulent, buoyant jets discharged to flowing
stratified ambients", Oak Ridge National Laboratory, ORNL-4685 (U).

238

DRAFT
Hgstrm, U. (1964), "An experimental study on atmospheric diffusion", Tellus, 16,
205-251 (U).
Hgstrm, U. (1987), "Non-dimensional wind and temperature profiles in the
atmospheric surface layer: a re-evaluation",
Boundary-Layer Met., 42, 55-78 (U).
Holtslag, A. A. (1984), "Estimates of diabatic wind speed profiles from near-surface
weather observations", Boundary-Layer Met., 29, 225-250 (U).
Holtslag, A. A. M. and A. P. van Ulden (1983), "A simple scheme for daytime estimates
of the surface fluxes from routine weather data",
J. Clim. & Appl. Met., 22, 517-529 (U).
Horn, H.-G., M. Maqua and H. Bonka (1988), "Nasse und trockene Ablagerung
radioaktiver Stoffe auf die Vegetation und den Erdboden", Schriftenreihe
Reaktorsicherheit und Strahlenschutz Der Bundesmeminister fr Umwelt,
Naturschultz und Reaktorsicherheit (U).
Incropera, F. P. and D. P. DeWitt (1990), "Fundamentals of Heat Transfer",
Wiley, 919pp (U).
Kasten, F. and G. Czeplak (1989), "Solar and terrestrial radiation dependent on the
amount and type of cloud", Solar Energy, 24, 177-189 (U).
Kato, H. and O. M. Phillips (1969), "On the penetration of a turbulent layer into a
stratified fluid", J. Fluid Mech., 37, 643-655 (U).
Kennard, E. H. (1938), "Kinetic Theory of Gases", McGraw-Hill, New York, 483pp (U).
Klebanoff, P. S. (1955), "Characteristics of turbulence in a boundary layer with zero
pressure gradient", Nat. Adv. Ctee. Aero., Wash., Report no.1247 (U).
Kukkonen, J., M. Kulmala and T. Wren (1994), "The homogeneous equilibrium
approximation in models of aerosol cloud dispersion",
Atmos. Env., 28, 2763-2776 (U).
Launder, B. E., G. J. Reece and W. Rodi (1975), "Progress in the development of a
Reynolds-stress turbulence closure", J. Fluid Mech., 68, 537-566 (U).
Leith, C. E. (1978), "Objective methods for weather prediction",
Ann. Rev. Fluid Mech., 10, 107-128 (U).

239

DRAFT
Lewellen, W. S. (1977), "Use of invariant modeling", Handbook of Turbulence,
ed. W. Frost and T. H. Moulden, Plenum Press, pp237-280 (U).
Lewellen, W. S. (1981), "Modeling the lowest 1km of the atmosphere", AGARD,
AGARD-AG-267 (U).
Lewellen, W. S. (1985), "Modeling turbulent exchange in forest canopies", The ForestAtmosphere Interaction, ed. B. A. Hutchison and B. B. Hicks, D. Reidel,
Dordrecht, pp481-499 (U).
Lewellen, W. S. and Y. P. Sheng (1980), "Modeling of dry deposition of SO2 and sulfate
aerosols", EPRI, EA-1452, Project 1306-1, 67pp (U).
Lewellen, W. S. and R. I. Sykes (1986), "Analysis of concentration fluctuations from
lidar observations of atmospheric plumes",
J. Clim. & Appl. Met., 25, 1145-1154 (U).
Macdonald, R. W., D. J. Hall, S. Walker and A. M. Spanton (1998), "Wind tunnel
measurements of windspeed within simulated urban arrays", Building
Research Establishment, Garston, UK, CR 243/98 (U).
MacFarlane, K., A. Prothero, J. S. Puttock, P. T. Roberts and H. W. M. Witlox (1990),
"Development and validation of atmospheric dispersion models for ideal
gases and hydrogen fluoride. Part 1; Technical Reference Manual", Shell
Research Ltd., TNER.90.015 (U).
Mantrom, D. D. and W. W. Haigh (1973), "Fireball Entrainment Study", Defense
Nuclear Agency, DNA 3248F (U).
Marshall, J. S. and W. M. Palmer (1948), "The distribution of raindrops with size",
J. Met., 5, 165-166 (U).
Mason, P. J. (1989), "Large-eddy simulation of the convective atmospheric boundary
layer", J. Atmos. Sci., 46, 1492-1516 (U).
Mason, P. J. and D. J. Thompson (1987), "Large-eddy simulations of the neutral-staticstability planetary boundary layer", Quart. J. Roy. Met. Soc.,
113, 413-444 (U).
McGahan, J. T. (1990), "NewFall", SAIC Internal Report, October 12,1990 (U).
Mellor, G. L. and H. J. Herring (1973), "A survey of mean turbulent field closure
models", AIAA J., 11, 590-599 (U).

240

DRAFT
Mikkelsen, T., S. E. Larsen and H. L. Pcseli (1987), "Diffusion of Gaussian puffs",
Quart. J. Roy. Met. Soc., 113, 81-105 (U).
Mosca, S., G. Graziani, W. Klug, R. Bellasio and R. Bianconi (1997), "ATMES-II Evaluation of long-range dispersion models using 1st ETEX release data:
Volume 1", JRC-Environment Institute (U).
Nastrom, G. D. and K. S. Gage (1985), "A climatology of atmospheric wavenumber
spectra of wind and temperature observed by commercial aircraft", J.
Atmos. Sci., 42, 950-960 (U).
Nieto, P. J. G., B. A. Garcia, J. M. F. Diaz and M. A. R. Brana (1994), "Parametric study
of selective removal of atmospheric aerosol by below-cloud scavenging",
Atmos. Env., 28, 2335-2342.
Nieuwstadt, F. T. M. (1981), "The steady-state height and resistance laws of the
nocturnal boundary layer: Theory compared with the Cabauw
observations", Boundary-Layer Met., 20, 3-17 (U).
Paine, R. J. (1987), "User's guide to the CTDM meteorological preprocessor (METPRO)
program", Atmos. Sci. Res. Lab., USEPA, Contract No. 68-02-3421 (U).
Paltridge, G. W. and C. M. R. Platt (1976), "Radiative Processes in Meteorology and
Climatology", Elsevier, 318pp (U).
Pennsyle, R. O. (1992), "NUSSE4: Improvements and extensions of the NUSSE3
model", U.S. Army Chemical Research, Development and Engineering
Center, Aberdeen Proving Ground, MD 21010-5423,
CRDEC-TR-325 (U).
Proctor, F. H. (1987), "The Terminal Area Simulation System. Volume I: Theoretical
Formulation", NASA Contractor Report 4046 (U).
Puttock, J. S. (1988), "A model for gravity-dominated dispersion of dense-gas clouds",
Stably stratified flows and dense gas dispersion, 233-260, ed.
J. S. Puttock, Clarendon Press, Oxford (U).
Raj, P. K. and J. A. Morris (1987), "Source characterization and heavy gas dispersion
models for reactive chemicals", U.S. Air Force, Systems Command, TMS
Report No. G-85-42 (U).

241

DRAFT
Raupach, M. R., J. J. Finnigan, and Y. Brunet (1996). Coherent eddies and turbulence in
vegetation canopies: the mixing-layer analogy.
Boundary-Layer Met. 78, 351-38. (U).
Saucier, R. (1987), "NUSSE3 model description", Chemical Research and Development
& Engineering Center, CRDEC-TR-87046 (U).
Sawford, B. L. (1982), "Comparison of some different approximations in the statistical
theory of relative dispersion", Quart. J. Roy. Met. Soc., 108, 191-206 (U).
Schmidt, H. and U. Schumann (1989), "Coherent structure of the convective boundary
layer derived from large-eddy simulations", J. Fluid Mech., 200, 511-562
(U).
Schumann, U. (1988), "Minimum friction velocity and heat transfer in the rough surface
layer of a convective boundary layer",
Boundary-Layer Met., 44, 311-326 (U).
Sehmel, G. A. (1973), "Particle eddy diffusivities and deposition velocities for isothermal
flow and smooth surfaces", Aerosol Sci., 4, 125-138 (U).
Seinfeld, J. H. (1986), "Atmospheric chemistry and physics of air pollution", Wiley, New
York, 738pp (U).
Slinn, W. G. N. (1982), "Predictions for particle deposition to vegetative canopies",
Atmos. Env., 16, 1785-1794 (U).
Spalart, P. R. (1989), "Theoretical and numerical study of a three-dimensional turbulent
boundary layer", J. Fluid Mech., 205, 319-340 (U).
Sparmacher, H., K. Flber and K. Bonka (1993), "Below-cloud scavenging of aerosol
particles: Particle-bound radionuclides-experimental",
Atmos. Env., 27A, 605-618 (U).
Sykes, R. I. (1984), "The variance in time-averaged samples from an intermittent plume",
Atmos. Env., 18, 121-123 (U).
Sykes, R. I., W. S. Lewellen and S. F. Parker (1984), "A turbulent-transport model for
concentration fluctuations and fluxes", J. Fluid Mech., 139, 193-218 (U).
Sykes, R. I., W. S. Lewellen and S. F. Parker (1986), "A Gaussian plume model of
atmospheric dispersion based on second-order closure",
J. Clim. & Appl. Met., 25, 322-331 (U).

242

DRAFT
Sykes, R. I., W. S. Lewellen, S. F. Parker and D. S. Henn (1988), "A hierarchy of
dynamic plume models incorporating uncertainty, Volume 4: Secondorder Closure Integrated Puff", EPRI, EPRI EA-6095 Volume 4,
Project 1616-28 (U).
Sykes, R. I. and D. S. Henn (1992), "An improved moment conservation method for the
advection-diffusion equation", J. Appl. Met., 31, 112-118 (U).
Sykes, R. I. and W. S. Lewellen (1992), "Review of potential models for UF6 dispersion",
Martin Marietta energy Systems, Inc., K/GDP/SAR-19 (U).
Sykes, R. I., D. S. Henn and C. P. Cerasoli (1993a), "Near surface dust: Late time
atmospheric effects", Defense Nuclear Agency, DNA-TR-93-175 (U).
Sykes, R. I., D. S. Henn and W. S. Lewellen (1993b), "Surface-layer description under
free-convection conditions", Quart. J. Roy. Met. Soc., 119, 409-421 (U).
Sykes, R. I., S. F. Parker, D. S. Henn and W. S. Lewellen (1993c), "Numerical simulation
of ANATEX tracer data using a turbulence closure model for long-range
dispersion", J. Appl. Met., 32, 929-947 (U).
Sykes, R. I., S. F. Parker, D. S. Henn and R. S. Gabruk (1996), "PC-SCIPUFF Version
0.2 Technical Documentation", Defense Nuclear Agency, DNA-TR-96-27
(U).
Sykes, R. I. and R. S. Gabruk (1997), "A second-order closure model for the effect of
averaging time on turbulent plume dispersion",
J. Appl. Met, 36, 165-184 (U).
Sykes, R. I., D. S. Henn and L. P. Santos (1998), "Description of chemical reaction /
radioactive decay and cloud shine in SCIPUFF", Defense Threat
Reduction Agency, DSWA-TR-98-10 (U).
Taylor, G. I. (1921), "Diffusion by continuous movements", Proc. Lond. Math. Soc.
(Ser. 2), 20, 196-211 (U).
Thomson, D. J. (1987), "Criteria for the selection of stochastic models of particle
trajectories in turbulent flows", J. Fluid Mech., 180, 529-556. (U)
Townsend, A. A. (1976), "The structure of turbulent shear flow", Cambridge University
Press, Cambridge, 429pp (U).

243

DRAFT
Ulden, A. P. van and A. A. M. Holtslag (1983), "The stability of the atmosphere surface
layer during nighttime", Sixth Symp. on Turbulence and Diffusion,
American Meteorological Society, Boston, MA, 257-260 (U).
Ulden, A. P. van and A. A. M. Holtslag (1985), "Estimation of Atmospheric Boundary
Layer Parameters for Diffusion Applications",
J. Climate Appl. Meteor., 24, 1196-1207 (U).
Venkatram, A. K. (1980), "Estimating the Monin-Obukhov length in the stable boundary
layer for dispersion calculations", Boundary-Layer Met., 19, 481-485 (U).
Venkatram, A. and J. C. Wyngaard (1988), "Lectures on air pollution modeling",
American Meteorological Society, Boston, MA (U).
Weil, J. C. (1988), "Dispersion in the convective boundary layer", Lectures on Air
Pollution Modeling, 167-227, ed. A. Venkatram and J. C. Wyngaard,
Amer. Met. Soc., Boston (U).
Weil, J. C., R. P. Lawson and A. R. Rodi (1993), "Relative dispersion of ice crystals in
seeded cumuli", J. Appl. Met., 32, 1055-1073 (U).
Wright, S. (1996), PDCALC User's Manual", STRATCOM Technical Report, July 1996
(S).
Wyngaard, J. C. (1975), "Modeling the planetary boundary layer - extension to the stable
case", Boundary- Layer Met., 9, 441-460 (U).
Wyngaard, J. C. (1985), "Structure of the Planetary Boundary Layer and Implications for
its Modeling", J. Climate Appl. Meteor., 24, 1131-1142 (U).
Yee, E. (1990), "The shape of the probability density function of short-term concentration
fluctuations of plumes in the atmospheric boundary layer",
Boundary-Layer Met., 51, 269-298 (U).
Zilitinkevich, S. S. (1972), "On the determination of the height of the Ekman boundary
layer", Boundary-Layer Met., 3, 141-145 (U).

244